JIAO Wenting

Exploring Risk Factors on Chinese A Share Stock
Market - in the Frame of Fama - French Factor Model

Wenting Jiao
To cite this version:

Wenting Jiao. Exploring Risk Factors on Chinese A Share Stock Market - in the Frame of Fama
- French Factor Model. Economies and finances. Université Rennes 1, 2017. English. <NNT :
2017REN1G013>. <tel-01685829>
HAL Id: tel-01685829

https://tel.archives-ouvertes.fr/tel-01685829
Submitted on 16 Jan 2018
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
ANNÉE 2017
THÈSE / UNIVERSITÉ DE RENNES 1

sous le sceau de l’Université Bretagne Loire
pour le grade de
DOCTEUR DE L’UNIVERSITÉ DE RENNES 1
Mention : Sciences de gestion
École doctorale EDGE

présentée par
Wenting JIAO
Préparée à l’unité de recherche UMR CNRS6211 CREM
Centre de Recherche en Économie et Management
Institut de Gestion de Rennes
Thèse soutenue à Rennes

Intitulé de la thèse le 21 septembre 2017
: Exploring Risk devant le jury composé de :
Factors on Chinese Khalid EL BADRAOUI

Professeur Habilité à Diriger des Recherches,
A-Share Stock Market Université Ibn Zohr (Maroc) / rapporteur

Franck MORAUX
– in the Frame of Professeur à l'Université de Rennes 1 / membre
Hélène RAINELLI-WEISS
Fama-French Factor Professeur à l'Université de Strasbourg / rapporteur
Jean-Jacques LILTI
Model Professeur à l'Université de Rennes 1 / directeur de
thèse
Acknowledgement
The journey of my research in IGR-IAE (Universitéde Rennes 1, France) will come to the
end soon. I am so lucky that I could study for my Ph.D. degree in France and in Rennes
with the help of so many excellent professors and colleagues. First of all, I would like to
thank my supervisor Mr. Jean-Jacques LILTI, I appreciate that you accept me as your
student, thanks for your patient guidance, encouragement, and advice that you have
provided throughout my research.
I would also like to express my deepest thank to Professor Franck MORAUX and Professor
Jean-Laurent VIVIANI, who teach me and help a lot during my research period in IGR.
You let me know that research is an interesting but hard work. I appreciate all other
professors who have helped me with my research.
I would also like to thank my colleagues and friends, who help me not only with my
research but also with my life in France. I cannot mention your names one by one here, but
I remember every moment we are together. Thank you for your help and encouragement,
you make my life in France happy and colorful.
Especially, I would like to express my gratitude and my love to my parents, my father
Xianfeng JIAO, my mother Sunhua JUAN. Without your love, support, and comprehension,
I cannot make it. I would also like to thank my boyfriend Jiali XU, thank for your help and
companion. I would like to express that how important you are in my life and how much I
love you!
To everyone who I love

Table of Contents
List of Tables ....................................................................................................................... vii
List of Figures ...................................................................................................................... xi
Abbreviations ..................................................................................................................... xiii
Résuméde la thèse en français ........................................................................................ xvii
Introduction .......................................................................................................................... 1
Background ......................................................................................................................... 1
Motivation .......................................................................................................................... 3
Objective and Structure of Dissertation ............................................................................. 7
Contributions and Discussion ............................................................................................. 9
Chapter 1 Fama-French Five-Factor Model vs. Fama-French Three-Factor Model in

China .................................................................................................................................... 11
1.1 Reviews of the main asset pricing theories and models ........................................ 12
1.1.1 The evolution of the asset pricing models ...................................................... 12
1.1.2 Reviews of Fama-French Three-Factor Model .............................................. 13
1.1.3 Reviews of Fama-French Five-Factor Model ................................................. 14
1.2 Retrospective of empirical work of Fama-French factor models .......................... 16
1.2.1 Overviews of the data and methodologies of FF factor models ..................... 16
1.2.1.1 Construction of Fama-French three factors and portfolios ..................... 16
1.2.1.2 Construction of profitability factor and investment factor ...................... 17
i
Table of Contents
1.2.2 Reviews of the empirical tests for Fama-French Three-Factor Model ........... 19
1.2.2.1 Empirical results of FF on some developed countries’ stock markets .... 19
1.2.2.2 Summary of empirical results in emerging Asian markets ..................... 22
1.2.2.3 Empirical tests in Chinese stock market ................................................. 25
1.2.3 Profitability and investment factors and Fama-French Five-Factor Model ... 28
1.3 Introduction of Chinese stock market .................................................................... 31
1.3.1 Background of Chinese stock market ............................................................. 31
1.3.2 Special features of Chinese stock market ....................................................... 36
1.4 Empirical Analysis and Results of FF3F Model .................................................... 38
1.4.1 Data and methodology .................................................................................... 38
1.4.1.1 Data ......................................................................................................... 38
1.4.1.2 Construction of Fama-French portfolios on Chinese market .................. 39
1.4.2 Empirical results on Chinese A-share stock market ....................................... 40
1.4.2.1 Summary statistics................................................................................... 40
1.4.2.2 Time-series regressions ........................................................................... 44
1.4.2.3 Cross-sectional results on Chinese stock market .................................... 50
1.4.3 Comparing with U.S. market (FF3F Model) .................................................. 52
1.5 Construction of profitability and investment factors and empirical results of FF5F
Model ............................................................................................................................... 53
1.5.1 Construction problems.................................................................................... 53
1.5.2 Empirical results of Fama-French Five-Factor Model on Chinese A-share
stock market .................................................................................................................. 56
1.5.2.1 Summary information of factors and portfolios ...................................... 56
1.5.2.2 Time-series regressions of FF5F model on Chinese A-share stock market
................................................................................................................. 58
1.5.3 Comparing with U.S. stock market (FF5F Model) ......................................... 63
1.6 Conclusions ............................................................................................................ 64
Chapter 2 Fama-French Factors and Innovations in State Variables on Chinese A-

Share Stock Market ............................................................................................................ 69
ii
Table of Contents
2.1 Economic explanation of Fama-French factors in the context of ICAPM ............ 70

2.1.1 ICAPM framework ......................................................................................... 70
2.1.2 What are the proper candidates for the state variables? ................................. 72
2.1.3 Evidence from outside of the U.S. market ..................................................... 78
2.2 Innovations in predictive variables and Vector Autoregressive approach ............ 81
2.2.1 State variables ................................................................................................. 81
2.2.2 Vector Autoregressive method ....................................................................... 83
2.2.3 Innovations in state variables ......................................................................... 84
2.3 Time-series evidence on Chinese A-share stock market ....................................... 85
2.3.1 Fama-French factors and innovations of state variables ................................ 86
2.3.1.1 Statistics description................................................................................ 86
2.3.1.2 Relation between FF factors and innovations of state variables ............. 88
2.3.2 Time-series regressions and results ................................................................ 90
2.3.2.1 Five comparative models and time-series regressions ............................ 91
2.3.2.2 Brief comparison and summary ............................................................ 104
2.4 Cross-sectional validation of five comparing models .......................................... 107
2.5 Conclusions .......................................................................................................... 113
Chapter 3 Distress Risk Factor and Stock Returns on Chinese A-Share Stock Market
.................................................................................................................................... 117
3.1 Financial distress risk ........................................................................................... 118
3.2 Measurement of financial distress risk ................................................................ 119
3.2.1 Accounting-based models ............................................................................ 120
3.2.1.1 Altman’s Z-score ................................................................................... 121
3.2.1.2 Ohlson’s O-score ................................................................................... 122
3.2.2 Market-based models .................................................................................... 124
3.2.2.1 Black-Scholes and Merton (BSM) option-pricing model ..................... 126
3.2.2.2 KMV model........................................................................................... 129
3.2.3 Hazard model ................................................................................................ 132
3.2.4 Comparing Accounting-based models and market-based models ................ 133
iii
Table of Contents
3.3 Financial distress risk and equity returns ............................................................. 135

3.3.1 Literature reviews on developed markets ..................................................... 135
3.3.1.1 Distress risk and expected stock returns ............................................... 135
3.3.1.2 Distress risk factor ................................................................................. 138
3.3.2 Empirical studies of distress risk on Chinese stock market ......................... 139
3.4 Construct distress risk factor and portfolios on Chinese stock market ................ 141
3.4.1 Data............................................................................................................... 141
3.4.1.1 Calculate default probability by O-score and DLI ................................ 142
3.4.1.2 Summary description of O-score and DLI ............................................ 143
3.4.2 Construction of distress factor and portfolios on Chinese market................ 145
3.4.2.1 Portfolios construction using O-score and DLI as proxy of distress risk ...
............................................................................................................... 145
3.4.2.2 Construction of distress risk factor........................................................ 146
3.5 Distress risk, size and B/P ratio of Chinese stock market ................................... 146
3.5.1 Distress risk and variation of stock returns .................................................. 147
3.5.2 Size effect ..................................................................................................... 149
3.5.3 Value effect................................................................................................... 152
3.6 Augmented four-factor model and empirical evidence ....................................... 155
3.6.1 Augmented four-factor model ...................................................................... 155
3.6.2 Time-series regressions analysis .................................................................. 156
3.6.2.1 Accounting-based distress risk factor and regression results ................ 157
3.6.2.2 Market-based distress risk factor and regression results ....................... 161
3.6.3 Cross-sectional regressions of the augmented four-factor model ................ 165
3.7 Conclusions .......................................................................................................... 167
General conclusions .......................................................................................................... 169
Bibliography ...................................................................................................................... 173
Appendix A Errors-in-Variables (EIV) problem and Shanken correction procedure ....... 193
Appendix B Performance of FF3F Model and FF5F Model on U.S. stock market .......... 195
iv
Table of Contents
Appendix C Time-series regression of six Size-B/P portfolios, six Size-OP portfolios and
six Size-Inv portfolios on FF3F Model (Chinese stock market) ........................................ 203
Appendix D Time-series regressions on single innovation of selected state variables in

addition to excess market return ......................................................................................... 205
Appendix E Derivation process of Merton’s default probability and distance-to-default 207
Appendix F Time-series and cross-sectional regressions of 18 Size-B/P-DR sorted

portfolios on FF3F model ................................................................................................... 209
Appendix G List of publications and conferences ............................................................ 213
v
List of Tables
Table 1.1 Summary of Chinese stock market 1990-2014 .................................................... 34

Table 1.2 Annual average available number of firms that have required data (2001-2014) 40
Table 1.3 Descriptive statistic of Fama-French six value-weighted Size-B/P portfolios
(period: July 2004- May 2015) ............................................................................................. 41
Table 1.4 Summary statistics of Fama-French three factors and correlation coefficients
among variables (July 2004- May 2015) .............................................................................. 42
Table 1.5 Descriptive statistic of Fama-French 25 Size-B/P portfolios (July 2004- May
2015) ..................................................................................................................................... 43
Table 1.6 Time-series regression of six value-weighted Size-B/P portfolios on Chinese A-
share stock market (July 2004- May 2015, 131 months) ..................................................... 45
Table 1.7 Time-series regressions of 25 value-weighted Size-B/P portfolios on FF3F Model,
Chinese A-share stock market (July 2004 - May 2015; 131 months) .................................. 46
Table 1.8 Cross-sectional regressions of six value-weighted Size-B/P portfolios and 25
value-weighted Size-B/P portfolios, Chinese A-share stock market (July 2004- May 2015)
.............................................................................................................................................. 51
Table 1.9 Annual firm numbers that have available data of OP and Inv (2004-2014) ........ 54
Table 1.10 Annual number of stocks in 25 Size-OP portfolios and 25 Size-Inv portfolios
(2010-2014) .......................................................................................................................... 55
Table 1.11 Annual number of stocks in six Size-OP portfolios and six Size-Inv portfolios 56
Table 1.12 Summary statistics of Fama-French five factors (July 2010-May 2015) ........... 57
Table 1.13 Average monthly excess returns for portfolios formed on Size-B/M, Size-OP
and Size-Inv (July 2010-May 2015, 59 months) .................................................................. 58
vii
List of Tables
Table 1.14 Time-series regressions of six value-weighted Size-B/P portfolios, Size-OP

portfolios and Size-Inv portfolios on FF5F Model, Chinese A-share stock market (July
2010 to May 2015, 59 months)............................................................................................. 59
Table 2.1 List of papers that document time variation of excess asset return and the state
variables they use.................................................................................................................. 82
Table 2.2 Summary statistics of FF factors, state variables and their innovations (December
2006-May 2015, 102 months) .............................................................................................. 86
Table 2.3 Correlation coefficients among risk factors ......................................................... 87
Table 2.4 Estimated coefficients of SMB and HML from risk factor regressions ............... 89
Table 2.5 Time-series regressions of Contemporaneous innovations of state variables on FF
factors ................................................................................................................................... 90
Table 2.6 Time-series regression on innovations of state variables and innovations of FF
factors (period: December 2006-May 2015) ........................................................................ 93
Table 2.7 Time-series regression on innovations of state variables and FF factors, period:
December 2006-May 2015 ................................................................................................... 96
Table 2.8 Time-series regression on innovations of state variables, period: December 2006-
May 2015 ............................................................................................................................ 100
Table 2.9 Time-series regression on innovations of FF factors, period: December 2006-May
2015 .................................................................................................................................... 102
Table 2.10 Time-series regression on FF3F model, period: December 2006-May 2015 ... 105
Table 2.11 Cross-sectional regressions of five comparative models.................................. 110
Table 3.1 Classification of financial distress prediction models ........................................ 119
Table 3.2 Annual available number of firms whose O-score and DLI can be calculated .. 144
Table 3.3 Characteristics of portfolios sorted by O-score and DLI (2005-2014)............... 148
Table 3.4 Size effect controlled by distress risk (O-score as proxy), 2005-2014 .............. 150
Table 3.5 Size effect controlled by distress risk (DLI as proxy), 2005-2014 .................... 151
Table 3.6 Value effect controlled by distress risk (O-score as proxy) ............................... 153
Table 3.7 Value effect controlled by distress risk (DLI as proxy) ..................................... 154
Table 3.8 Summary statistics of four factors: market factor, SMB, HML and distress risk
factor (July 2005 to May 2015) .......................................................................................... 156
viii
List of Tables
Table 3.9 Summary statistics of 18 (2x3x3) Size-B/P-O-score sorted portfolios (July 2005-
May 2015) .......................................................................................................................... 158
Table 3.10 Time-Series Regressions of 18 Size-B/P-O-score sorted portfolios on
augmented four-factor model (July 2005-May 2015) ........................................................ 159
Table 3.11 Summary statistics of 18 (2x3x3) Size-B/P-DLI sorted portfolios .................. 162
Table 3.12 Time-Series Regressions of 18 Size-B/P-DLI sorted portfolios on augmented
four factor model (Period: July 2005-May 2015) ............................................................... 163
Table 3.13 Cross-sectional regressions of 18 Size-B/P-DR sorted portfolios on augmented
four-factor model, on FF3F Model, and on market factor and DRF (July 2005-May 2015,
119 months) ........................................................................................................................ 166
ix
List of Figures
Figure 1.1 SHASHR Index and SZASHR Index (1992-2015)............................................. 33

Figure 1.2 SHASHR Index and NYSE Composite Index (1992-2015) ............................... 33
Figure 1.3 Total listed firms and that of SSE and SZSE, listed stocks of A-share stock
market (1990-2014) .............................................................................................................. 35
Figure 1.4 Total market capitalization and that of A-share and B-share stock market
(unit:100 million yuan; 1992-2014) ..................................................................................... 36
Figure 1.5 Loadings of 25 Size-B/P portfolios on FF3F ...................................................... 48
Figure 1.6 Annual number of firms that has available data of OP and Inv .......................... 54
Figure 3.1 Equity as a European call option on the firm .................................................... 127
Figure 3.2 Annual aggregate O-score (2005-2014) ............................................................ 144
Figure 3.3 Annual aggregate DLI (2005-2014) .................................................................. 145
xi
Abbreviations
3M T-Bill Rate: Three-Month Treasury Bill Rate
ANNA: Artificial Nerve Network Analysis
B/M: Book-to-market equity
B/P: Book-to-price
BS: Black-Scholes
BSM: Black-Scholes and Merton
CDS: Credit default swaps
Chinese A-share: CNAS
Cross-sectional regression: CSR
CSRC: China Securities Regulatory Commission
DD: Distance-to-default
DEF: Default spread
DIV: Dividend yield
DLI: Default likelihood indicator
DP: Default (distress) probability
DR: Distress (default) risk
DRF: Distress risk factor
E/P: Earning-to-price
xiii
Abbreviations
EDF: Expected default frequency
FF: Fama-French
FF3F Model: Fama-French Three-Factor Model
FF5F Model: Fama-French Five-Factor Model
FM: Fama-MacBeth
GBM: Geometric Brownian Motion
GEB: Growth Enterprise Board
GDP: Gross Domestic Product
GNP: Gross National Product
ICC: Implied cost of capital
IDEF: Innovations of default spread
IDIV: Innovations of dividend yield
IHML: Innovations of HML
Inv: Investment
IRF: Innovations of one-month deposit rate
ISMB: Innovations of SMB
ITERM: Innovations of term spread
LTD: Long-term debts
MDA: Multiple discriminant analysis
OP: Operating profitability
PT: Particular transfer
RF: Risk-free Rate (one-month deposit rate)
RMB: RenMinBi (Chinese ‘yuan’)
SBPD: Size-B/P-DLI
xiv
Abbreviations
SBPO: Size-B/P-O-score
SDF: Stochastic Discount Factor
SH: Shanken
SHASHR Index: Shanghai Stock Exchange A-Share Index
SME: Small Medium Enterprise Board
SSE: Shanghai Stock Exchange
ST: Special treatment
STD: Short-term debts
SV: Survival rate
SZASHR Index: Shenzhen Stock Exchange A-Share Index
SZSE: Shenzhen Stock Exchange
t-stats: t-statistics
TERM: Term spread
Time-series regression: TSR
VAR: Vector-autoregressive
xv
Résuméde la thèse en français
Introduction générale
Les modèles de facteurs ont étéles modèles dominants dans le domaine de prix de l’actif
depuis des décennies depuis l’apparition du ‘Capital Asset Pricing Model’ (CAPM) de
Sharpe (1964), Lintner (1965) et Black (1972). Le CAPM d’origine peut être considéré
comme le modèle de facteur àindice unique et facile àeffectuéet àinterpréter; cependant,
il est critiqué pour ses hypothèses et le manque du pouvoir explicatif des anomalies du
marchéboursier. D’innombrables chercheurs se sont consacrés àdévelopper un modèle qui
remédie aux défauts de CAPM. Les adversaires de CAPM qui voient son manquement du
pouvoir explicatif affirment qu’il doit y avoir d’autres facteurs en dehors de la rentabilité
excédentaires du marchéqui dérivent la rentabilité de l’actif. Les modèles d’évaluation des
actifs qui ont plus d’un facteur sont appelés modèles multifactoriels.
L’un des modèles multifactoriels fondamentaux est ‘Arbitrage Pricing Theory’ (APT)
introduite par (Ross, 1976). Dans cette théorie, la plupart des hypothèses sous-jacentes au
CAPM sont assouplies. L’APT suppose qu’il y a n facteurs qui causer la rentabilité de
l’actif s’écartent systématiquement de leurs valeurs espérées, mais la théorie ne précise pas
àquel point le nombre n, ni identifie les facteurs.
Fama et French (1993) déduit un modèle empirique purement àtrois facteurs qui inclut un
facteur liéàla taille de l’entreprise (SMB) et un facteur liéau ratio book-to-market (HML)
de l’entreprise, en plus de bêta du marchéde CAPM, c’est le Modèle àTrois Facteurs de
Fama-French (FF3F) célèbre. Ils fournissent de fortes preuves empiriques que le modèle à
trois facteurs capture la plupart des variations la rentabilité de l’actif en coupe dans le
marché d’action des États-Unis. En outre, la source ouverte sur le site de Kenneth R.
French facilite également les recherches. Il est également intéressant de noter que le modèle
de Fama-French (FF) est l’une des principales raisons pour lesquelles le Professeur Fama
reçoit le Prix Nobel de Economie 2013. Depuis l’apparition du Modèle FF3F, une énorme
quantitéde travail a étéeffectuéselon le Modèle FF3F, par exemple, chercher de nouveaux
xvii
facteurs ou examiner comment le Modèle FF3F fait sur marchés des actions différents
partout dans le monde. Actuellement, plus de chercheurs se tournent vers la recherche de la
fondation économique en arrière les facteurs ou les applications en pratique.
Les modèles de facteurs, en particulier le CAPM et le Modèle FF3F, occupent pas

seulement un rôle pivot dans le développement de la théorie d’évaluation des actifs, mais
contribuent également àla pratique du marchéet àl’analyse des investissements. Ils sont
largement utilisés par les gestionnaires de portefeuille, les investisseurs institutionnels, les
gestionnaires financiers et les investisseurs individuels, à tel que la prédiction de la
rentabilitéd’actif, la sélection de titres, la construction de portefeuille et contrôle du risque,
quantifier l’exposition au risque d’un portefeuille par rapport à un indice de référence,
mesurer la performance, et l’évaluation du gestionnaire de portefeuille.
L’utilisation des modèles de facteurs comme une base pour les recherches est devenue
standard dans la littérature financière. La preuve que les facteurs FF sont largement utilisés
peut être trouvée dans les journaux principaux dans le domaine financier, ‘The Journal of
Finance’ et ‘Journal of Financial Economics’. Ces deux journaux de premier ordre sont
classés dans les revues A-étoile (l’évaluation le plus élevée). Ils sont habituellement classés
dans les deux ou trois revues financières les plus importantes du monde. Plus récemment,
Fama et French (2015a) proposent un modèle àcinq facteurs visant àcapturer la taille, la
valeur, la rentabilité et l’investissement dans le rendement d’action sur le marchéboursier
américain dans le ‘Journal of Financial Economics’; et fournir le test international du
modèle àcinq facteurs (Fama et French, 2017).
En outre, plus preuves empiriques ou pratiques que les modèles de facteurs sont largement
utilisés peuvent être trouvés dans ‘The Journal of Portfolio Management’ (JPM), qui est
une des revues qui propose des recherches en pointe sur l’allocation d’actifs, la mesure de
la performance, les tendances du marché, la gestion des risques, l’optimisation de
portefeuille, etc. Les articles publiés par JPM proviennent pas seulement des chercheurs les
plus renommés, mais aussi des praticiens renommés. En particulier, JPM a publié un
numéro spécial de 2017 pour célébrer ses 40 ans, qui contient les recherches les plus
récentes relatives aux facteurs de risque ou aux modèles des facteurs. Trois articles (Bass et
al., 2017; Cocoma et al., 2017; et Bender et Samanta, 2017) prend l’objet d’allocation de
portefeuille fondée sur des facteurs. Quatre articles sont basés sur l’investissement fondé
sur les facteurs. Deux autres articles sont sur les méthodologies pour construire les
portefeuilles de facteurs: Amenc et al. (2017) fournissent deux méthodes de construction de
portefeuilles multifactoriels, tandis que Liu (2017) propose un portefeuille de quintile avec
un nouveau cadre de construction de facteurs.
Nous énumérons ci-dessous plusieurs applications de modèles de facteurs en pratique:
xviii
- Gestion de portefeuille
Le pouvoir d’un modèle multi-facteurs est que compte tenu des facteurs des risques et des
sensibilités aux facteurs des risques, le profil d’exposition au risque d’un portefeuille peut
être quantifié et contrôlé. Un gestionnaire de portefeuille peut également analyser son
risque actuel et comprendre la taille et l’emplacement de ses paris. D’autre part, les
modèles de facteurs peuvent être utilisés pour décomposer le risque du portefeuille en
fonction de l’exposition aux facteurs communs, et pour évaluer la part de la rentabilitéd’un
portefeuille attribuable àchaque exposition aux facteurs communs. En utilisant un modèle
de multifactoriel et un modèle d’optimisation, on peut construire un portefeuille présentant
le minimum de risque actif par rapport à son indice de référence pour un nombre donné
d’actifs détenus. Similairement, un gestionnaire de portefeuille peut construire un
portefeuille qui s’incline vers un facteur spécifié et ce portefeuille n’a aucune exposition
active matérielle à aucun autre facteur. En plus, les modèles multifactoriel de risque
permettent à un gestionnaire et à un client d’évaluer les performances potentielles d’un
portefeuille ou d’une stratégie de négociation par rapport àun indice de référence.
- Estimation de rentabilitéexigéet évaluation de l’équité
Les modèles de facteurs sont largement utilisés pour estimer le taux de rentabilitédu capital
exigé, compte tenu du risque pertinent. Si les investisseurs s’attendent à une rentabilité
d’action particulier supérieur àune rentabilitéexigé, l’action est sous-évalué; en revanche,
si la rentabilitéespérée est inférieur au taux de rentabilitéexigé, l’action est surévalué. Le
taux de rentabilité exigé est une composante dans plusieurs des métriques et des calculs
utilisés dans finance d’entreprise et évaluation de l’équité.
L’évaluation de l’équitéest une partie centrale dans des nombreuses activités, telles que la
sélection des titres, l’analyse des actions, en déduisant des attentes du marché, l’évaluation
des événements d’entreprise et l’évaluation des entreprises privées. Les gestionnaires des
actifs effectuent les évaluations parce que leur objectif principal est identifier les titres mal
cotés (sous-évalués ou surévalués). Les banques d’investissement qui jouent un rôle
intermédiaire dans l’événements d’entreprise, par exemple dans les fusions et acquisitions,
procèdent également à l’évaluation de l’entreprise cible. Un analyste d’action évalue les
actions qu’elle (ou il) suite (suit) pour donner des recommandations (acheter, tenir, vendre).
Les analystes ou les chercheurs utilisent souvent des modèles d’évaluation pour extraire les
attentes du marché. Par exemple, un analyste peut inscrire le prix du marché et le taux
d’intérêt, le dividende et d’autres facteurs dans un modèle d’évaluation de l’équitéafin de
déterminer quel taux de croissance le marchéimplique.
- La pratique de la finance d’entreprise et le coût de l’estimation de l’équité
xix
Le coût du capital est la rentabilitéexigénécessaire pour réaliser un projet de budgétisation

en capital. Il comprend le coût de la dette et le coût de l’équité. Les modèles de facteurs
sont généralement utilisés pour mesurer le coût de l’équité(COE). COE est la rentabilité
que les actionnaires ont besoin pour leur investissement dans une entreprise. Le COE d’une
entreprise représente la compensation que la demande du marchéen échange de posséder
l’actif et portant le risque de propriété. Les actionnaires s’attendent à obtenir une certain
rentabilitésur leurs participations dans une entreprise, le taux de rentabilitéexigépar les
actionnaires est un coût au point de vue de l’entreprise. Le coût de l’équité est
essentiellement ce qu’il coûte à l’entreprise de maintenir un prix d’action qui est
théoriquement satisfaisant pour les investisseurs. Sur cette base, la méthode le plus
couramment acceptée pour calculer le coût de l’équité provient du CAPM et le Modèle
FF3F.
- Investissement àbase de facteur
Investissement à base de facteur est devenu une partie largement discutée du canon
d’investissement d’aujourd’hui. Les chercheurs d’investissement utilisent des modèles
multifactoriels des risques pour exécuter des back-tests contrôlés sur les stratégies
d’investissement futures. Pour cela, leurs besoins sont semblables àceux des gestionnaires
de portefeuille. Ils doivent mettre en œuvre des stratégies optimales sur les données
historiques et comprendre la performance ultérieure de ces stratégies. Les chercheurs
peuvent utiliser des back-tests pour améliorer leurs stratégies. Ils peuvent également utiliser
l’analyse de performance et la caractérisation des risques de portefeuille pour améliorer leur
compréhension des paris qu’ils testent. La preuve que l’investissement factoriel est en
vogue peut également être trouvée dans le dernier numéro spécial (2017) de Journal of
Portfolio Management. Deux articles (Dimson et al., 2017 et Kim et al., 2017) se
concentrer sur l’investissement basésur les facteurs, et deux articles (Podkaminer, 2017 et
Alford et Rakhlin, 2017) sur la bêta intelligente qui est une stratégie d’investissement
factorielle populaire actuellement. Par exemple, MSCI a créé une famille d’indices de
facteurs (index bêta intelligents) qui permettent d’accéder à six facteurs solides: valeur,
taille faible, volatilitéfaible, rendement élevé, qualitéet élan.
Une débauche de recherches et l’utilisation pratique largement des modèles de facteurs sont
les deux motivations de cette thèse. En plus, la croissance d’économique significative en
Chine au cours des dernières décennies a été universellement reconnue, la Chine a
maintenant la deuxième grande économie au monde, et elle est le plus grand pays en
développement. Le marchéboursier chinois, crééen 1990, est un représentant des marchés
émergents avec une histoire relativement courte. Son développement ainsi que son
immaturitéont attirél’attention des chercheurs, et ont apportés la question si les modèles
d'évaluation des actifs tels que le modèle de FF s’appliquent également à ses marchés
xx
domestiques. Par conséquent, la popularitédes modèles de facteurs et le marchéboursier

émergent stimulent l’intention pour l’étude des modèles factoriels sur le marché boursier
chinois. On peut dire que ceci encore une autre étude des modèles de facteurs sur la base du
Modèle FF3F, mais ce que nous voulons souligner, c’est que le modèle FF3F est devenu la
pierre angulaire des études des facteurs, la recherche sur les modèles de facteurs a étéun
sujet de recherche chaud pendant des décennies, et il continuera d’être un thème clé du
domaine d’évaluation des actifs àl’avenir.
Le travail présenté dans cette thèse contient trois chapitres, explorant principalement les
facteurs de risque sur le marchéboursier chinois sur la base du Modèle FF3F. Le succès du
Modèle FF3F sur le marché boursier des États-Unis stimule les études sur l’applicabilité
dans d’autres marchés développés ou émergents partout dans le monde, l’applicabilitédu
Modèle FF3F sur le marchéboursier chinois a également étéun sujet d’actualitédepuis des
années. Cette thèse commence par un nouveau test du Modèle FF3F àl’aide des données du
marché boursier chinois A-share, afin de répondre à la question si le Modèle FF3F
s’applique au marchéboursier chinois A-share. Les travaux suivants se déroulent àpartir de
Modèle FF3F.
Le chapitre 1 ré-enquête sur le Modèle FF3F et explore le dernier Modèle àCinq Facteurs
de Fama-French (FF5F) sur le marché boursier chinois A-share, vise à examiner
l’applicabilité des deux modèles en Chine. En outre, nous comparons les résultats de
régression des deux modèles pour vérifier si Modèle FF5F fonctionne mieux que le Modèle
FF3F pour capturer la variation des rentabilités des actions. Il est important de noter que
nous prenons en compte plusieurs caractéristiques particulières du marchéboursier chinois
qui ne sont pas négligeables pour bénéficier de résultats de recherche plus fiables.
Pour mettre en œuvre l’enquête, nous construisons les cinq facteurs: rentabilités
excédentaire du marché, facteur de taille SMB, facteur de valeur HML, facteur de
rentabilité RMW et facteur d’investissement CMA en utilisant des données du marché
boursier chinois A-share. L’approche à deux étapes de Fama-MacBeth (1973) est appliqué
pour les régressions séries chronologiques et les régressions transversales. Étant donnéque
l’approche àdeux étapes de Fama-MacBeth (FM ci-dessous) a causéle problème d'erreurs
sur les variables (EIV), la procédure de correction proposée par Shanken (1992) est
appliquée.
Bien que la grande réussite du modèle de facteur FF, c’est aussi l’un des modèles d’
évaluation des actifs qui sont les plus controversés, principalement pour ses considérations
purement empiriques et le manque de fondements théoriques. Le succès du Modèle FF3F
stimule la poursuite des bases économiques des facteurs FF. Au chapitre 2, nous procédons
à la recherche conformément à l’une des explications dominantes, basée sur des
xxi
opportunités d’investissement variant dans le temps dans le cadre de l’ICAPM. Nous

examinons si les facteurs FF sont proxys des innovations de variables d’état sélectionnées
qui décrivent les opportunités d’investissement futures sur le marché boursier chinois A-
share. Nous choisissons quatre variables économiques selon Petkova (2006): rendement en
dividende agrégée, taux de T-bonds en un mois, l’écart de terme et l’écart de défaut, afin de
modéliser deux aspects de l’opportunitéd’investissement set- la courbe de rendement et la
distribution conditionnelle des rentabilités des actifs.
Pour extraire les termes de l’innovation, le processus de Vecteur Autoregressif (VAR) a été
utilisé. Et comme Campbell (1996) souligne que’il est difficile d’interpréter les résultats
d’estimation pour un modèle de facteur VAR, sauf si les facteurs soient orthogonalisés et
mis àl’échelle d’une certaine manière", nous orthogonalisons les innovations des variables
d’état sur la rentabilitéexcédentaire du marché, qui est le premier élément du vecteur dans
le processus VAR. L’approche àdeux étapes de Fama-MacBeth est ensuite mise en place
pour les régressions séries chronologiques et les régressions transversales, et les résultats de
la régression transversale sont ajustés pour le problème de l’EIV.
Dans le chapitre 3, nous choisissons le facteur de risque détresse comme facteur augmenté
du Modèle FF3F original parmi les facteurs de risque qui ont étédécouverts. Étant donné
que le risque détresse de l’entreprise est un indicateur important de la performance d’une
entreprise, et est également l’une des caractéristiques les plus concernées par les
investisseurs et les entreprises. De plus, le risque de détresse financière a étéprouvéliéaux
rentabilités des actions étroitement et le point de vue de l’effet de la valeur est considéré
comme l’effet du risque de détresse proposépar la littérature séries chronologique.
L’objectif principal est de tester si les facteurs FF sont des procurations pour le risque de
détresse sur le marchéboursier chinois pendant la période de juillet 2005 àmai 2015. Sinon,
si le modèle à quatre facteurs augmenté explique la rentabilité d’action mieux que le
Modèle FF3F. En particulier, nous appliquons le modèle àbase de comptabilité(O-score
d’Ohlson) et le modèle de marché (DLI des Vassalou et Xing) pour mesurer la détresse
financière, afin d’identifier si les méthodes différentes qui prédisent du risque de détresse
sont importantes pour les résultats empiriques. En outre, la relation entre la rentabilité
d’action et le risque de détresse, et si l’effet de taille et l’effet de valeur sont liés au risque
de détresse ont également étéexaminés sur le marchéboursier chinois.
xxii
Chapitre 1
Modèle à Cinq-Facteurs de Fama-French versus Modèle à
Trois-Facteurs de Fama-French en Chine
Contexte et objectif
Fama et French (1993) proposent le Modèle FF3F et démontrent que le facteur de taille et
le facteur de valeur ont effectué réussi à expliquer les rentabilités des actions en plus du
bêta du marché. Dernier, Fama et French (2015a) proposent un modèle à cinq facteurs
incluant le facteur rentabilité et le facteur d’investissement en plus des trois facteurs
original de FF, et ce modèle se comportent mieux que le Modèle FF3F.
Le Modèle FF3F et le Modèle FF5F sont:
Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ei ,t (1)
Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ri RMW  ciCMA  ei ,t (2)
Il est évident que le modèle àcinq facteurs comporte deux autres facteurs que le modèle à
trois facteurs, RMW et CMA. RMW est le facteur liéàla rentabilitéde l’entreprise, ce qui
correspond à la différence de rentabilité entre les portefeuilles qui ont les rentabilités
robustes et les rentabilitéfaible. Le CMA est liée àl’investissement, lequel est la différence
de rentabilité entre les portefeuilles qui ont les investissements conservatifs et les
portefeuilles qui ont les investissements agressifs.
Ce chapitre ré-examine l’applicabilité du Modèle FF3F en construisant des facteurs FF sur

le marchéboursier chinois A-share (séries chronologique et transversales), compte tenu de
plusieurs caractéristiques spéciales du marchéboursier chinois. De plus, ce chapitre étudie
également le dernier Modèle FF5F en utilisant les données du marchéboursier chinois, afin
d’examiner si les facteurs de rentabilité et d’investissement ont un pouvoir explicatif
supplémentaire. En outre, nous comparons deux modèles pour voir si le Modèle FF5F
fonctionne mieux que le Modèle FF3F sur le marché boursier chinois A-share. Les
comparaisons des résultats empiriques entre le marché boursier chinois et le marché
boursier américain sont également exécutés.
Caractéristiques spécials du marchéboursier chinois
xxiii
Les littératures empiriques suggèrent que le marché chinois présente des caractéristiques
spécials, et il est inévitable de considérer ces caractéristiques spécials si les chercheurs
veulent avoir des résultats plus précis en Chine. Dans notre étude, nous résumons trois
caractéristiques principales qui sont également les plus fréquemment employées par les
littératures: (1) utiliser des actions négociables, pas toutes les actions (y compris les actions
négociables et non-négociables) pour pondérer la rentabilité d’action en valeur, (2) inclure
les entreprises de ‘Petit Moyen Enterprise Planche’ et ‘Croissance Enterprise Planche’ pour
déterminer les points de ruptures pour le facteur de taille, et (3) calculer le ratio book-to-
price (B/P) au lieu du ratio book-to-market (B/M) puisque la segmentation du marché
boursier chinois .
Données et méthodologie
Nous incluons toutes les entreprises sur le marché boursier chinois A-share (marché
boursier A-share de Shanghai et marchéboursier A-share de Shenzhen), à l’exclusion des
entreprises financières et des entreprises ayant des valeurs des B/P négatives; en plus, une
entreprise est éliminée si les informations pertinentes manquent dans un mois ou une
période donné. La période de recherche pour le Modèle FF3F est de juillet 2004 àmai 2015
(131 mois), et le Modèle FF5F est de juillet 2010 àmai 2015 (59 mois).
L’approche en deux étapes de Fama-MacBeth (FM) est appliquée pour les régressions, la
première étape est la série chronologique en utilisant l’OLS. La deuxième étape est la
régression transversale telle qu’indiquée dans l’équation (1), où ils utilisent les bêtas
estimées obtenues de la première étape des régressions chronologiques, comme les
variables indépendantes dans les régressions transversales. Et réglez les rentabilités des
mêmes portefeuilles (que dans la première étape) sur ces bêtas estimées pour une période
de temps fixe pour déterminer la prime de risque pour chaque facteur.
Ri ,t  R f  i   M bî   SMB ˆsi   HML hî  i ,t (3)
Étant donné que l’approche en deux étapes de FM cause le problème d'erreurs sur les
variables (EIV) classiques, la procédure de correction proposée par Shanken (1992) est
ensuite appliquée aux écarts types des estimateurs de régression transversale.
Construction de portefeuilles et de facteurs
À la fin de juin de chaque année t, tous les actions sont triés en deux groupes de taille, Petit
et Grand, le point de rupture est la médiane de la capitalisation boursière totale. Nous
xxiv
classons par B/P au lieu du B/M à la fin de chaque décembre de l’année t-1 en trois groupes:
Bas, Moyen et Haut cela dépend des points de rupture des 30e et 70e percentiles de A-share.
Enfin, l’intersection de ces six groupes fait les six portefeuilles, qui restent les mêmes de
juillet de l’année t à juin de l’année t+1, et les portefeuilles sont réformés en juillet de
l’année t+1.
La médiane de la
capitalisation boursière
Petit Bas Grand Bas

30e percentil B/P
Petit Moyen Grand Moyen
70e percentil B/P
Petit Haut Grand Haut
Les facteurs SMB et HML est construit comme suit:
1
SMB  ( Petit Bas+Petit Moyen+Petit Haut)
3 (4)
1
 ( Grand Bas+Grand Moyen+Grand Haut )
3
1 1
HML  ( Petit Haut+Grand Haut )  ( Petit Bas+Grand Bas ) (5)
2 2
Similairement, la construction de portefeuilles sur la rentabilité opérationnelle (OP) et

l’investissement (Inv) est similaire en celle des portefeuilles sur le ratio B/P. À la fin de
chaque mois de juin, les entreprises sont triées en trois portefeuilles OP basésur des points
ruptures des 30e et 70e percentiles de l’OP, et les trois portefeuilles d’investissement sont
formés à l’aide de points rupture: les 30e et 70e percentiles d’Inv. Ensuite, les facteurs
RMW et CMA sont les suivants:
1 1
RMW  ( Petit Robuste+Grand Robuste)  ( Petit Faible+Grand Faible ) (6)
2 2
1
CMA  ( Petit Conservatif+Grand Conservatif)
2 (7)
1
 ( Petit Agressif+Grand Agressif )
2
xxv
Nous construisons trois séries de portefeuilles: six portefeuilles Taille-B/P pondérées en

valeurs, six portefeuilles Taille-OP pondérées en valeurs et six portefeuilles Taille-Inv
pondérées en valeurs, en tant que variables dépendantes pour effectuer les régressions.
Résultats et conclusions
Le Modèle FF3F peut expliquer la majoritédes variations chronologiques des rentabilités

des actions, compte tenu des caractéristiques spécials du marchéboursier chinois. Au cours
des périodes de juillet 2004 à mai 2015, il existe une prime de taille positif dans la
rentabilité d’actions sur le marché boursier chinois A-share, toutefois, nous constatons le
manque de prime de valeur. Le bêta du marché est capable d’expliquer la variation
transversale des rentabilités moyens pour les 25 portefeuilles pondérés en valeur (avec des
t-statistiques SH ajustés négatifs).
Pour tous les trois séries de portefeuilles que nous avons construits dans la recherche du
Modèle FF5F, il existe un effet de taille; tandis qu’il existe un effet de valeur dans les
portefeuilles Taille-B/P, l’effet de rentabilité dans les portefeuilles Taille-OP et l’effet
d’investissement dans les portefeuilles Taille-Inv. Les coefficients sur RMW ne sont
significatives que dans la serie des portefeuilles qui formés àpartir de taille et OP. En ce
qui concerne le facteur CMA, les coefficients significatives sont concentrées dans les
groupes OP ou Inv extrêmes, tels que le groupe OP faible, le groupe OP robuste, les
groupes Inv agressifs et conservatif. Cependant, pour les portefeuilles Taille-B/P, les
coefficients significatifs de CMA sont relativement dispersifs.
Nous comparons la performance entre le Modèle FF3F et le Modèle FF5F au cours de la

période de recherche, si le Modèle FF5F fonctionne mieux que le Modèle FF3F sur le
marchéboursier chinois A-share n’est pas très clair. Le pouvoir explicatif du Modèle FF5F
est différent parmi les séries de portefeuilles différents. Par rapport au Modèle FF3F, la
présence de facteurs de rentabilitéet d’investissement ne capturer pas plus de variations des
rendabilités des actions espérés que le modèle à trois facteurs, à l'exception des six
portefeuilles pondérées en valeurs qui formés à partir de la taille et de la rentabilité
opérationnelle, bien que l’amélioration soit limitée.
Nous comparons également performances des Modèles FF entre le marchéboursier chinois

et le marché boursier américain. Les résultats révèlent que le Modèle FF3F et le Modèle
FF5F expliquent la variation de séries chronologiques des rendabilités meilleure sur le
marchéboursier américain par rapport au marchéboursier chinois. En ce qui concerne les
deux facteurs supplémentaires, le facteur de rentabilitéRMW et le facteur d’investissement
CMA, sont capables de capturer partiellement les variations de séries chronologiques des
xxvi
rendabilités des trois séries portefeuilles sur le marchéboursier américain, alors que sur le
marchéboursier chinois, le facteur de rentabilitésemble être un facteur explicatif seulement
pour les six portefeuilles Taille-OP.
Chapitre 2
Fama-French Facteurs et Innovations des Variables d'état sur le
MarchéBoursier Chinois
Objectif
Le Modèle FF3F a réaliséun énorme succès empirique depuis son apparition, donc il est
considérécomme l'un des modèles les plus controversés d'évaluation actif. Cependant, les
facteurs sont construits empiriquement .donc ils manquent de fondements théoriques. En
particulier, leurs liens économiques avec le risque systématique ne sont pas clairs. Par
conséquent, la performance impressionnante du Modèle FF3F a suscité des recherches
nombreuses qui tentent de fournir une interprétation économique claire des facteurs HML
et SMB.
Les liens économiques sous-jacents des facteurs FF sont plutôt controversés. Parmi des
nombreuses des explications concurrentes pour le succès du Modèle FF3F, suivant que
Petkova (2006), dans ce chapitre, nous concentrons sur celui basésur des opportunités des
investissements qui variant dans le temps et dans le contexte d’Intertemporal Capital Asset
Pricing Model (ICAPM ci-après) de Merton (1973a).
Dans ce chapitre, nous examinons si les innovations des quatre variables d'état, rendement
du dividende agrégée, taux de T-bonds en un mois, l'écart de terme et l'écart de défaut, sont
capables de capturer les rentabilités excédentaires des actions dans les séries
chronologiques et séries transversales. Nous examinons également si FF Facteurs SMB et
HML sont les proxys pour les innovations de variables d'état sélectionnées, qui décrivent
les opportunités d'investissement futures sur le marchéboursier chinois A-share.
xxvii
Le même sérier des variables d'état que Petkova (2006) est choisi en plus des facteurs FF
dans notre test empirique: rendement de dividende agrégée (DIV), l'écart de terme (TERM),
l’écart de défaut (DEF) et taux de T-bonds en un mois (RF), qui sont parmi les variables
économiques qui sont utilisées les plus communes dans les littératures. La période de
recherche est de décembre 2006 àmai 2015 (102 mois).
Selon ICAPM, seul le composant inattendu de la variable d'état devrait commander une
prime de risque. Le composant inattendu est normalement appelé innovations. Au lieu
d'utiliser directement les variables d'état pour l’implémentation empirique de l'ICAPM,
Campbell (1996) suggère d'utiliser des innovations dans telles variables d'état pour prévoir
les changements dans le sérier des opportunités d'investissement futures. Pour dériver les
termes de l'innovation, nous appliquons la méthode vecteur autorégressif (VAR).
 Rm,t   Rm,t 1 
 DIV   DIV 
 t
  t 1

 TERM t TERM t 1 
   
 DEFt  = A  DEFt 1  + ut (8)
 RFt   RFt 1 
   
 SMBt   SMBt 1 
 HMLt   HMLt 1 
Où ‘A’ est une matrice 7 × 7, et représente un vecteur d'innovations de 7 × 1 pour chaque

élément. Il y a six innovations correspondant au rendement du dividende, au taux de T-
bonds en un mois, à l'écart de terme (TERM), à l’écart de défaut, au SMB et au HML, qui
sont extraits de et designés IDIV, IRF, ITERM, IDEF, ISMB et IHML.
Les innovations peuvent être exprimées dans le cadre de l’ICAPM:
Ri ,t  R f ,t   i ,t   i ,m  RM ,t  R f ,t     i , k  tk   ei ,t
K
(9)
1
où, tk est l’innovation de l’état variable k au temps t.
Ensuite, les innovations des variables d'état sont séparément orthogonalisées au facteur du
marché. Précisément, l'innovation de la rentabilitéexcédentaire du marchéreste inchangée,
l'innovation du rendement de dividende IDIV est orthogonaliséàla rentabilitéexcédentaire
du marché. De même, IRF est orthogonalisée au facteur du marché. Les autres innovations
des variables d'état ITERM, IDEF, ISMB et IHML sont traitées par la même manière.
xxviii
Puis l’approche à deux étapes de Fama-MacBeth est implémentée pour effectuer les
régressions pour cinq modèles comparatifs, et la correction de Shanken est également
effectuée pour ajuster le problème EIV dans les régressions transversales.
Cinq modèles comparatifs
Modèle 1
Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF

(10)
 i ,IRF IRF  i ,ISMB ISMB  i ,IHML IHML  ei ,t
Modèle 2

(11)
 i ,IRF IRF  i ,SMB SMB  i ,HML HML  ei ,t
Modèle 3

(12)
 i ,IRF IRF  ei ,t
Modèle 4
Ri ,t  R f  i  i ,m ( RM ,t  R f )  i ,ISMB ISMB  i ,IHML IHML  ei ,t (13)
Modèle 5
Ri ,t  R f  i  i ,m ( RM ,t  R f )  i ,SMB SMB  i ,HML HML  ei ,t (14)
Le modèle 1 représente les régressions chronologique sur la rentabilité excédentaire du

marché, les innovations des variables d'état et les innovations des facteurs FF SMB et HML
(ISMB et IHML). Le modèle 2 représente les régressions chronologiques sur la rentabilité
excédentaire du marché, les innovations des variables d'état et les facteurs FF initiaux SMB
et HML. Les variables indépendantes dans le modèle 3 sont les rentabilités excédentaire du
marché et quatre innovations des variables d'état. Dans le modèle 4, les variables
indépendantes sont les rentabilités excédentaire du marché et les innovations de SMB et
HML. Le modèle 5 n'est que le Modèle FF3F original.
Nous réalisons les séries chronologiques suivant les cinq modèles ci-dessus et réalisons
également les régressions transversales correspondant aux cinq modèles comparatifs.
xxix
Nous pouvons conclure que les facteurs FF (ou les innovations des facteurs FF) expliquent
bien la variation des séries chronologiques des rentabilités attendus des actions, avec ou
sans les innovations des variables d'état dans le modèle. Lorsque les quatre innovations des
variables d'état sont respectivement régressées, seulement IDIV a un pouvoir explicatif
pour la rentabilité moyenne; en présence de facteurs FF (ou des innovations des facteurs
FF), IDIV perd sa capacitéàcapturer les variations des rentabilités chronologiques attendus.
La présence des quatre innovations de variables d'état qui prévoient des opportunités
d'investissement futures ne expulse pas les facteurs FF. L'information contenue dans
l'innovation des rendements de dividende IDIV semble totalement capturée par la
combinaison du marchébêta et le SMB (ou ISMB). Bien que le modèle comporte àla fois
des facteurs FF et des innovations de variables d'état (comme les facteurs de risque)
effectue légèrement meilleurs que le Modèle FF3F original (compte tenu de R-carré
moyenné ajusté), les FF facteurs ont pu jouer un rôle limité en capturant d'opportunités
d'investissement alternatives représentées par les innovations des variables d'état.
Chapitre 3
Facteur de Risque de Détresse et Rentabilités des Actions sur le
MarchéBoursier Chinois
Objectif
L'étude du chapitre 2 est basée sur l'une des trois théories spécifiques les plus discutées:
l'exposition aux changements dans les variables économiques dans le contexte de l'ICAPM
(les deux autres théories spécifiques sont le risque de détresse et l'exposition asymétrique
aux conditions économiques). Les résultats démontrent que les facteurs FF ne représenter
pas les innovations des quatre variables sélectionnées sur le marché boursier chinois A-
share pendant la période de décembre 2006 àmai 2015. Une autre explication théorique, le
risque de détresse, est étudiédans ce chapitre. Puisque une explication de la performance
des rentabilitépersistants des hauts B/M actions est le risque de détresse financière.
xxx
Ce chapitre étudie la relation entre les rentabilites des actions et le risque de détresse, et
aussi examine si l’effet de taille et l’effet de valeur sont liés au risque de détresse sur le
marchéboursier chinois A-share. De plus, nous explorons un modèle àquatre facteurs en
ajoutant un facteur de risque de détresse en plus des trois facteurs FF afin d'examiner si
facteurs FF sont des proxys de facteurs de risque de détresse. Nous examinons si différentes
méthodes de construction des facteurs entraînent des résultats différents.
La mesure de risque de détresse financière
Parmi les méthodes de prévision du risque de détresse, l'utilisation de ratios comptables et

d'informations sur le marché sont deux classifications dominantes. Les modèles qui
utilisent des ratios comptables pour estimer le risque de détresse s'appellent les modèles
basés sur la comptabilité, par exemple, le ‘Z-score’ de Altman (1968) et le ‘O-score’ de
Ohlson (1980). Les modèles qui utilisent l'information du marchépour estimer le risque de
détresse sont les modèles basés sur le marché, représentés par le modèle de ‘option-pricing’
de Merton (1974) et le modèle KMV de Moody.
Pour mesurer le risque de détresse financière, deux modèles de prédiction dominants sont
mis en place sur le marchéboursier chinois: O-score de Ohlson (1980) (modèle basésur la
comptabilité)
O  1.32  0.407 log( SIZE )  6.03(TLTA)  1.43(WCTA)  0.076 (CLCA)

(15)
1.72(OENEG )  2.37 ( NITA)  1.83( FUTL)  0.285( INTWO)  0.521(CHIN )
et le modèle de de ‘option-pricing’ de Merton (1974) (modèle basésur le marché)
  VA,t   1 2 
 ln        A T 
DP=N    Xt   2   (16)
t
 A T 
 
 
Vassalou et Xing (2004) soulignent que leur probabilitédéfaut (DP) estimée en utilisant du
modèle de Merton n'est pas le DP réel, au lieu de cela, le DP calculébasésur la distribution
empirique du modèle KMV de Moody est le DP actuel. Ainsi, ils appellent leur mesure de
DP comme l’indicateur de probabilitépar défaut (DLI).
Afin d'étudier si les trois facteurs FF sont des proxys pour le risque de détresse, un modèle
augmentéàquatre facteurs, qui comprend le facteur de marché(bêta du marché), le facteur
xxxi
de taille (SMB), le facteur de valeur (HML) et le facteur de risque de détresse (DRF), est
examiné:
Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  di DRF  i ,t (17)
Si toutes les informations dans SMB et HML sont liées au risque de détresse financière, il
est probable de trouver que SMB et HML perdent toute leur capacité à expliquer les
rentabilités des équités. en présence de DRF.
Dans ce chapitre, les données de recherche proviennent d’une période de juillet 2005 àmai
2015 (119 mois). Dans la recherche des deux méthodes, les entreprises financières, dont la
structure capitale se distingue de celle des ordinaires, est exclu. Les entreprises ayant un
ratio B/P négatif sont également retiréde l'échantillon. Puisque cet recherche se concentre
sur les entreprises saines, les entreprises qui ont qualifié ‘Traitement spécial’ et ‘Transfert
particulier’ en Chine sont également hors de notre considération.
Pour procéder la recherche et pour effectuer la comparaison, tout d'abord,la probabilitépar

défaut est calculée en appliquant la formule de O-score d'Ohlson (mesure basée sur la
comptabilité) et une procédure itérative complexe est utilisé pour résoudre le modèle de
‘option-pricing’ afin de calculer le DLI (mesure basée sur le marché). Ces probabilités par
défaut sont ensuite utilisées comme un critère principal pour la distinction du risque de
détresse des entreprises.
Les proxys du risque de détresse que nous utilisons sont les DP calculés par O-score et DLI.
À la fin de chaque mois de juin de l'année t, nous classons les actions selon leur O-score
(ou DLI) de fin d'année t-1 en trois groupes, désignés comme O1, O2 et O3 (ou DLI1,
DLI2, DLI3) de bas àhaut des probabilités défauts. Les points des reptures sont 30% et 70%
des percentiles du O-score (ou DLI) d'échantillonnage. Les portefeuilles restent immuables
de juillet de l'année t àjuin de l'année t+1, et il sont réformés àla fin de juin de l'année t+1.
Les données sont traité comme ça pour toute la période de recherche. Le portefeuille
imitant DRF est le facteur de risque de détresse, qui mesure la différence de rentabilité
entre le portefeuille DP haut (O3 ou DLI3) et le portefeuille DP bas (O1 ou DLI1).
Les comparaisons sont effectuées entre les résultats obtenus à partir des deux modèles
différents (le modèle basésur la comptabilitéet le modèle basésur le marché). Ensuite, de
la même manière que les chapitres précédents, l'approche àdeux étapes de Fama-MacBeth
et la méthode ajustée EIV de Shanken sont appliquépour réaliser les régressions.
xxxii
Tout d'abord, nous trouvons qu'il existe un effet de taille fort, mais aucun effet de valeur.
L'effet de taille et le manque d'effet de valeur sont robustes lorsqu'ils sont contrôlés par le
risque de détresse ainsi que sur l'ensemble de l'échantillon du marchéboursier chinois. Et
les effets sont également solides pour les proxys différentes du risque de détresse.
À partir des régressions chronologique, nous trouvons qu'il existe un effet de risque de
détresse significatif, les portefeuilles qui ont le risque de détresse plus haut portent
meilleure rentabilités sur le marchéboursier chinois A-share, le pouvoir explicatif de DRF
existe dans le top extrême (plus haut DR) et le bas (plus bas DR) groupes de portefeuille.
Le pouvoir explicatif des trois facteurs de FF n'a pas de changement significatif avec ou
sans DRF est présentédans le modèle, nous pouvons donc conclure que les facteurs FF ne
peuvent pas représenter DRF, le DRF explique plutôt les rentabilités chronologiques
moyennes en combinaison avec des facteurs FF sur le marché boursier chinois A-share.
Toutefois, le pouvoir explicatif supplémentaire de DRF est limité.
Par ailleurs, en comparant les résultats de régression selon en utilisant O-score et DLI
comme les proxys du risque de détresse, la performance du facteur de risque de détresse
basésur le DLI semble légèrement meilleure que celui basésur le O-score.
Nous fournissons des preuves àpartir des régressions transversales que les coefficients sur
DRF ne sont jamais un déterminant important des rentabilités moyennes, peu importe le
DRF est régresséavec le marchébêta ou des trois facteurs de FF. D’autre part, peu importe
le DRF est construit en utilisant O-score ou DLI comme le proxy, il existe une prime de
taille robuste (prime de marchérobuste lorsque DR est estiméen utilisant O-score). Ni le
facteur de valeur HML ni le facteur de risque de détresse DRF est capable de capturer les
variations transversale sur les rentabilités moyennes des actions.
Les études empiriques montrent que les facteurs FF ne perdent pas leur pouvoir explicatif
en présence de DRF. Nous pourrions conclure que, bien que DRF est un facteur évalué
pour déterminer les rentabilités chronologiques moyennes, il n'est pas le cas pour
déterminer les rentabilités transversales moyennes. Les facteurs FF ne peuvent pas
représenter DRF dans la section transversale du marchéboursier chinois A-share.
xxxiii
Conclusions générales
L'objectif principal de cette thèse est d'explorer les facteurs de risque et les modèles des
facteurs sur le marchéboursier chinois A-share sur le contexte du modèle facteur de Fama-
French.
Les résultats principaux de cette thèse sont présentés comme suit:
Tout d'abord, l'applicabilitédu Modèle FF3F est ré-examinépendant la période de juillet

2004 à mai 2015, en tenant compte de plusieurs caractéristiques spécials du marché
boursier chinois. Les résultats empiriques montrent que le Modèle FF3F peut expliquer la
majorité des variations de séries chronologiques des rentabilités des actions chinoises A-
share, en utilisant la valeur marchande négociable pour pondérer les portefeuilles et la
capitalisation boursière totale pour décider le point de rupture de taille, et le ratio B/P au
lieu du ratio de B/M. Les résultats des régressions transversales sont conformes aux la
plupart des études antérieures sur le marchéboursier chinois, le marchébêta et le SMB sont
des déterminants importants pour expliquer la variation transversale des rentabilités
moyennes des actions. Au cours de la période d'échantillonnage, il existe une prime de
marchénégative et une prime de taille positive sur le marchéboursier chinois A-share, mais
aucune prime de valeur est trouvé. Ces résultats sont robustes avec l'ajustement EIV, et
indépendants de l'intervalle de recherche.
L'applicabilité du dernier Modèle FF5F sur le marché boursier chinois A-part est aussi
étudiéau cours de la période de juillet 2010 àmai 2015. Afin de réaliser cette étude, trois
séries de portefeuilles, six portefeuilles Taille-B/P pondérés en valeur, six portefeuilles
Taille-OP pondérés en valeur et six portefeuilles Taille-Inv pondérés en valeur, sont
construisons. Pour tous les trois séries de portefeuilles, les trois facteurs originaux – le
facteur de marché, le facteur de taille et le facteur de valeur – possèdent toujours un
pouvoir explicatif des séries chronologiques pour les rentabilités excédentaires attendus, en
présence de facteurs de rentabilitéet d'investissement. Il existe toujours un effet de taille
dans tous les trois séries des portefeuilles et les rentabilités excédentaires sont négativement
liés à la taille de l'entreprise; il existe l’effet de valeur dans les portefeuilles Taille-B/P,
l'effet de rentabilité dans les portefeuilles Taille-OP et l'effet d'investissement dans les
portefeuilles Taille-Inv. Le pouvoir explicatif du facteur RMW n'existe que dans les six
portefeuilles Taille-OP. Le facteur CMA n’explique les rentabilités moyennes des
portefeuilles que dans les groupes OP ou Inv extrêmes, par exemple le groupe OP faible, le
groupe OP robuste, les groupes Inv conservatifs et agressifs. Tandis que les coefficients
significatives sur CMA pour les portefeuilles Taille-B/P sont relativement dispersives.
xxxiv
D’après la comparaison des performances des modèles FF3F et FF5F en présence de

facteurs de rentabilité et d'investissement, le Modèle FF5F ne peut pas capturer plus de
variations des rentabilités des actions attendus que le Modèle FF3F, à l'exception des six
portefeuilles pondérées en valeurs qui formés à partir de la taille et de la rentabilité
opérationnelle (même si l'amélioration est limitée).
Dans le chapitre 2, nous examinons si les facteurs FF SMB et HML, sont les proxys
d’innovations de variables d'état sélectionnées (rendement de dividende agrégée, taux de T-
bonds en un mois, l’écart de terme et l’écart de défaut) qui décrivent, pendant la période
recherche, les opportunités futures d'investissement sur le marchéboursier chinois A-share.
Les régressions chronologiques et les régressions transversales sont réalisées sur cinq
modèles comparatifs. Les résultats empiriques indiquent que les facteurs FF ne perdent pas
leur pouvoir explicatif, avec ou sans la présence des innovations des quatre variables d’état
sélectionnés, à la fois dans les examens de séries chronologiques et les examens
transversaux. Les résultats des régressions transversales révèlent également qu'il existe les
primes significatives de risque de marché et de taille.. Ils montrent que l'information
contenue dans l'innovation du rendements en dividende (IDIV) est totalement capturée par
la combinaison du marchébêta et du facteur de taille Les facteurs FF ont pu jouer un rôle
limité dans la captation d'opportunités d'investissement alternatives représentées par les
innovations des quatre variables d'état sélectionnées.
Basé sur les découvertes, dans le chapitre 3, nous étudions si les facteurs FF sont des
proxys de facteurs de risque de détresse et si différentes méthodes de construction des
facteurs entraînent des résultats différents. Les résultats empiriques suggèrent qu'il n'y a pas
de preuve significative que les facteurs FF représentent un risque de détresse sur le marché
boursier chinois. La présence de DRF a peu d'effet sur le pouvoir explicatif de la série
chronologique des trois facteurs FF, alors que le DRF, combinéavec les trois facteurs FF,
peut partiellement expliquer de la rentabilité excédentaires moyenne des séries
chronologiques. En comparant les résultats des régressions des séries chronologiques à
partir de deux méthodes différentes, la performance du facteur de risque de détresse basé
sur le DLI semble légèrement meilleure que celui basésur le O-score. Cependant, le facteur
de risque de détresse n'est pas un déterminant important des rentabilités transversales
moyennes, et les facteurs FF ne peuvent pas représenter le facteur de risque de détresse
dans la section transversale du marchéboursier chinois A-share.
De plus, les études sur les facteurs de risque en Chine présentés dans cette thèse ont
également fourni de nouvelles implications en pratique sur le marchéboursier chinois:
Tout d'abord, en tenant compte de plusieurs caractéristiques spécials du marché boursier

chinois, les résultats indiquent que le facteur de marchéet le facteur de taille SMB sont des
xxxv
déterminants importants des rentabilités transversale des actions. Et l'existence de prime de

taille et le manque de de prime de valeur sur le marché boursier chinois A-share sont
indépendantes de la période de recherche. Basé sur les résultats du Modèle FF3F sur le
marché boursier chinois, tels que les gestionnaires d'actifs, ils peuvent créer des
portefeuilles qui inclinent vers le facteur de taille SMB plutôt que le facteur de valeur HML
afin de gagner la taille de prime. De plus, les gestionnaires d'actifs ou les investisseurs
individuels sont capable d'évaluer la performance potentielle d'un portefeuille par rapport
au Modèle FF3F comme l’indice de référence.
Ensuite, selon les résultats de l'examen du Modèle FF5F sur le marchéboursier chinois A-
share, nous concluons que les facteurs de rentabilité et d'investissement ont un pouvoir
explicatif supplémentaire limitéet en comparaison avec le modèle original àtrois facteurs,
le Modèle FF5F n'a pas d'amélioration significative àexpliquer la rentabilitéexcédentaire
moyenne. Nos résultats est incompatible avec ceux sur le marché boursier américain.
Similairement, si les investisseurs veulent investir sur le marché boursier chinois, il est
préférable de sélectionner les portefeuilles construits en fonction de Modèle FF3F au lieu
du Modèle FF5F. Cependant, si les investisseurs investissent sur le marché boursier
américain, il est judicieux de choisir les portefeuilles construits selon le Modèle FF5F,
puisque le Modèle FF5F est meilleur que le Modèle FF3F sur le marchéboursier américain.
En plus, pendant l’étude d’explication économique des facteurs FF sur le marchéboursier

chinois, nous trouvons que les facteurs FF ne perdent pas le pouvoir explicatif en présence
des innovations de variables d'état sélectionnées dans les régressions chronologiques et les
régressions transversales sur le marchéboursier chinois A-share. Comparer avec le modèle
original FF3F, la présence d'innovations des variables d'état ne capture pas plus de variation
de rentabilité moyenne. Les résultats indiquent que sur le marché boursier chinois, la
construction de portefeuilles incline vers les innovations des quatre variables économiques
ne peut pas générer une prime de risque supplémentaire. Le Modèle FF3F semble être le
meilleur choix en pratique àce jour sur le marchéboursier chinois.
Enfin, dans chapitre 3, il est démontréque les facteurs FF ne sont pas les proxys du facteur
du risque de détresse sur le marché boursier chinois, le facteur de risque de détresse,
combinéavec les trois facteurs FF, explique plutôt la rentabilitémoyenne excédentaire des
séries chronologiques. Le modèle augmenté à quatre facteurs explique la variation
chronologique des rentabilités moyennes des action meilleur que celui du Modèle FF3F.
Dans ce cas, par exemple, si les portefeuilles sont construisent, en plus des facteurs FF,
incline vers le facteur de risque de détresse imitant, on peut s'attendre àplus de rentabilité
moyennes que les portefeuilles qui sont construisent uniquement basésur des facteurs FF
initiaux. Cependant, le bénéfice supplémentaire provient du facteur de risque de détresse est
limité. Par ailleurs, les résultats peuvent également être réalisés par les entreprises pour
xxxvi
estimer le coût de l'équité; par les investisseurs pour évaluer la valeur inhérente des actions
et pour prendre leurs décisions.
Dans l'ensemble, afin de suivre le rythme des réformes de la Chine et de la mondialisation

de son économie, le marché boursier chinois a connu un développement rapide et des
réformes institutionnelles globales. Compte tenu des caractéristiques spécials, les
rentabilités des actifs et ses déterminants pourraient être différents entre le marchéboursier
chinois et ceux des marchés boursiers développés (tels que le marché américain et le
marché européen). Nos recherches actuelles ne peut pas de trouver l'explication
économique du succès des facteurs FF sur le marché boursier chinois A-share, nous
proposons donc d'examiner les facteurs de risque qui présentent les caractéristiques
particulières du marché boursier chinois ou d'autres variables économiques relatif aux
rentabilités des actions dans les recherches futures.
xxxvii
Introduction
Background
The factor models have been the dominant models in the asset pricing field for decades
since the come up of Capital Asset Pricing Model (CAPM) of Sharpe (1964), Lintner (1965)
and Black (1972), a model which was aimed to estimate required return on assets under the
main assumptions of Markowitz's portfolio theory. Even before the introduction of the
popular CAPM, Markowitz (1959) proposed the use of a single-factor model to explain
security returns. Sometimes referred to as index models, factor models often rely on the use
of factor analysis to identify factors that influence asset returns.
The original CAPM can be regarded as the single-index factor model and has advantages of
being easy to implement and interpret; however, it is criticized by many analysts for its
assumptions and lack of explanatory power of stock market anomalies. Countless
researchers have devoted themselves to develop a model that overcome the defects of
CAPM, and the opponents of CAPM who see its lack of explanatory power, claim that
there must be other factor(s) apart from excess market returns that derive the asset return.
The asset pricing models that have more than one factor are called multifactor models.
Multifactor models can be divided into three types - fundamental, macroeconomic and
statistical factor models. Connor (1995) gives a comprehensive overview of these three
types of factor models: Fundamental factor models use the returns to portfolios associated
with observed security attributes such as dividend yield, the book-to-market ratio, and
industry identifiers; macroeconomic factor models use observable economic time series,
such as inflation and interest rates, as measures of the pervasive shocks to security returns;
statistical factor models derive their pervasive factors from factor analysis of the panel data
set of security returns. The fundamental and macroeconomic factor models are models with
known factors, while the statistical factor cannot directly observable or easily interpreted in
contrast to known factors, they are also called latent factors. Latent factors are estimated by
applying a statistical methodology known as factor analysis and the model generated is
referred to as a factor model by statisticians, which is beyond the scope of this study.
The known factors normally used are observable economic, company and industry
attributes and market data as “descriptors” (e.g. GDP growth, interest rate, price/earnings
ratios, book-to-price ratios, estimated earnings growth, trading activity, and the business
cycle). However, the descriptors are not factors, instead, they are the candidates for risk
factors and are selected in terms of their ability to explain historical stock returns. The
1
Introduction
descriptors selected are then used to construct risk factors (economic or political factors or
industry factors or country factors) and referred to as risk indices. The most popular factors
today- Value, Growth, Size, Momentum- have been studied for decades as part of the
academic asset pricing literature and the practitioner risk factor modeling research.
The underlying premise of these multifactor models is that certain common factors may be
identified which capture the types of risk that are rewarded over investment horizons.
Factor models are widely used by investors to link the risk exposures of the assets to a set
of factors. In most of these models, the market index developed in the context of the CAPM
is one of the factors, representing "market risk."
One of the fundamental multifactor models is the Arbitrage Pricing Theory (APT)
introduced by Ross (1976). In this theory, most of the assumptions underlying the CAPM
are relaxed. APT assumes that there are n factors that cause asset returns to systematically
deviate from their expected values, however, the theory does not specify how large the
number n is, nor does it identify what the factors are.
Another epoch-making paper was published by Fama and French (FF) in 1993, they deduce
a three-factor model which includes a factor related to firm size (SMB) and a factor related
to firm’s book-to-market equity ratio (HML) in addition to market beta of CAPM, that is
the well-known Fama-French Three-Factor Model (FF3F Model) 1 . They provide strong
empirical evidence that the three-factor model captures most variation of cross-sectional
excess stock returns in U.S. stock market. And the new factors have been demonstrated to
add significantly to the prior understanding of returns based on the standard CAPM.
Furthermore, the open source on the website of Kenneth R. French also facilitates
following researchers. It is also worth noting that the FF model is one of the key reasons for
Professor Fama being awarded the 2013 Nobel Prize in Economics. Since the came up of
FF3F Model, an enormous amount of work has been implemented based on FF3F Model,
such as focusing on finding new factors or examining how FF3F Model is doing on
different stock markets all around the world. Nowadays, more researchers turn to seeking
the economic underpinning behind the factors or the applications in practice.
Why the heat of factor models continues for decades and no any trace of decrease until now?
There is a reason, the factor models, especially CAPM and FF3F Model, do not only
occupy a pivotal place in the development of the asset pricing theory, but also make great
1
Other multifactor models such as Burmeister-Ibbotson-Roll-Ross macroeconomic factor model
developed by Burmeister et al. (1994), MSCI Barra fundamental factor model, Barclay Group Inc.
bond factor model. In addition, the factor models developed by Haugen and Baker (1996a), and
Chen and Zhang (2010) are also proven improvements on the CAPM-inspired single-index market
model in explaining the cross-section of expected stock returns.
2
Introduction
contributes to market practice and investment analysis. The factor models are implemented
such as to select securities, construct a portfolio and control risk, quantify the risk exposure
of a portfolio relative to a benchmark index, measure performance, and in portfolio
manager evaluation.
Motivation
The factor models, are mentioned in a host of academic papers, and are widely used by
portfolio managers, institutional investors, financial managers, and individual investors to
predict asset returns; especially the original CAPM and FF3F Model which have each been
recognized by the Nobel Committee, are among the most implemented factor models in the
practice. For instance, a portfolio manager evaluates the impact of a series of broad factors
on the performances of various securities; in this sense, a reliable factor model provides a
valuable tool to assist portfolio managers with the identification of pervasive factors that
affect large members of securities. Researchers use multifactor models to back-test and
fine-tune strategies, and traders use these models to control investment risk over short
horizons, etc.
The use of the factor models as a basis for research has become standard in the finance
literature. Evidence of the fact that the FF factors are widely used can be found in the
leading journals in the field, The Journal of Finance and Journal of Financial Economics.
Both of these leading journals are ranked as A-star journals (the highest possible rating).
They are commonly ranked as being in the top two or three finance journals worldwide.
Most recently, Fama and French (2015a) propose a five-factor model aims at capturing the
size, value, profitability, and investment patterns in average excess stock returns on U.S.
stock market in the Journal of Financial Economics; and provide the international test of the
five-factor model Fama and French (2017).
Furthermore, more empirical or practical evidence that factor models are widely used can
be found in The Journal of Portfolio Management (JPM), which is one of the journals that
offers cutting-edge research on asset allocation, performance measurement, market trends,
risk management, portfolio optimization, and so on. Articles issued by JPM are not only
from most renowned researchers but also from the renowned practitioners. Particularly,
JPM issued a special issue of 2017 to celebrate its 40 years, which contains the most recent
researches related to risk factors or factor models. Three articles (Bass et al. (2017),
Cocoma et al. (2017), and Bender and Samanta (2017)) take the subject of factor-based
portfolio allocation; four articles are on the basis of factor-based investing. Another two
articles tend to the methodologies in studies of construction factor portfolios: Amenc et al.
3
Introduction
(2017) provide two methods of building multifactor portfolios, while Liu (2017) proposes a
pure quintile portfolio with new factor construction framework.
We list below several applications of factor models in practice:
- Portfolio management
The power of a multifactor model is that given the risk factors and the risk factor
sensitivities, a portfolio’s risk exposure profile can be quantified and controlled. A portfolio
manager can also analyze their current risk and understand the size and location of their
bets. On the other hand, factor models can be used to decompose portfolio risk according to
common factor exposure and to evaluate how much of a portfolio's return was attributable
to each common factor exposure. Consider a portfolio with more than 100 assets, instead of
estimating the variance-covariance matrix of its assets, it only necessary to estimate the
portfolio’s factor exposures and the variance-covariance matrix of the factors, a
computationally much easier task. Using a multifactor risk model and an optimization
model, a portfolio that has the minimum active risk relative to its benchmark for a given
number of assets held can be constructed. Similarly, a portfolio manager can construct a
portfolio that tilts towards a specified factor, and has no material active exposure to any
other factor. Furthermore, multifactor risk models allow a manager and a client to assess
the potential performance of a portfolio or trading strategy relative to a benchmark. In sum,
multifactor models can be used to analyze portfolio risk, construct portfolios that optimally
trade off risk with expected returns, and analyze skill and value added associated with past
returns. Consequently, the factor models offer a useful extension of the CAPM and the
APT because they advance our understanding of how key factors influence portfolio risk
and return.
- Required return estimation and equity valuation
The factor models are used largely to estimate the required return on equity, given the
relevant risk. That is, the model has the purpose of estimating normal risk-adjusted returns.
Estimating normal risk-adjusted returns are also required for a number of purposes
including setting regulated returns and for the purposes of determining whether a portfolio
or trading strategy shows out-performance on a risk-adjusted basis. The use of the factor
models, especially FF3F Model, as a basis for estimating required returns has become
standard in the finance literature. For example, the issues of February 2014 and December
2013 of the Journal of Finance feature five articles that use the FF factors for the purposes
of estimating required returns. The volume (2014) of the Journal of Financial Economics
features four articles that use the FF factors for the purposes of estimating required returns.
Indeed, the use of the Fama-French factors, for the purpose of estimating the required
4
Introduction
return on equity, is so widespread in the academic literature, its use as a measure of normal
returns has become a matter of course.
If investors expect a return of a particular stock higher than required return, then the stock
is undervalued; in contrast, if expected return is less than required rate of return, then the
stock is overvalued. The required rate of return is a component in many of the metrics and
calculations used in corporate finance and equity valuation.
The main goal of equity valuation is to estimate its inherent value. For investors, they are
concerned with the value of their investment, they all want to make sure that the product
they buy or sell is worth the value (price) they pay or receive. Inherent value is the “true” or
“real” value of an asset that is obtained and supported by rational, hypothetical, unbiased
mathematical models that consider all relevant factors that drive the value of an asset.
However, the inherent value is unobservable, so the best we can do is to estimate it using
asset pricing models. Such, the factor models are one category of the models that are
supposed to estimate the inherent value of common equities2. Inherent value can differ from
market value, in which case stock becomes overvalued or undervalued depending on
whether the inherent value is lower or higher than market value.
Equity valuation is a central part in many activities such as stock selection, stock analysis,
inferring market expectations, evaluation of corporate events (mergers and acquisitions,
divestitures, spin-offs), and private business valuation. Equity portfolio managers do equity
valuation of companies they consider including in a portfolio. Active managers, in
particular, do valuation because their primary goal is to identify mispriced (undervalued or
overvalued) securities. Investment banks who play an intermediary role in corporate events,
for example in mergers and acquisitions, also carry out the valuation of the target company.
A stock analyst evaluates the stocks she follows to give recommendations (buy, hold, sell).
Analysts or researchers also frequently use valuation models to extract (infer) market
expectations. For example, an analyst may input the current market prices and interest rates,
dividends, and other factors, into an equity valuation model in order to find what growth
rate the market implies.
- Corporate finance practice and the cost of equity estimation
The cost of capital is the required return necessary to make a capital budgeting project and
includes the cost of debt and the cost of equity. The factor models are generally used to
measure the cost of equity (COE). COE is the return that stockholders require for their
investment in a company. A firm's COE represents the compensation that the market
2
The dividend discount model and the free cash flow to equity model are also two well-known
models that are supposed to estimate inherent value of common equities.
5
Introduction
demands in exchange for owning the asset and bearing the risk of ownership. Common
shareholders expect to obtain a certain return on their equity investment in a company, the
equity holders' required rate of return is a cost from the company's perspective. The cost of
equity is basically what it costs the company to maintain a share price that is theoretically
satisfactory to investors. On this basis, the most commonly accepted method for calculating
the cost of equity comes from CAPM and FF3F Model.
If you ask participants involved in capital allocation decisions at large corporations “What
model do you use to estimate the cost of equity capital?” the most common answer is the
CAPM. However, when you evaluate the responses in detail you observe a large proportion
of respondents make adjustments to the CAPM estimate of the cost of equity capital to
account for other risk factors. What is proposed in the use of the FF model is an objective
measurement of risk that is consistent with the empirical evidence on stock returns.3
In 1999, a survey was conducted amongst 392 representatives of large corporations in the
U.S. (Graham and Harvey, 2001). In 2002, the same set of questions was posed to 313
representatives of large corporations in the U.K., the Netherlands, Germany, and France
(Brounen et al., 2004). In response to the question, “How do you determine your firm’s cost
of equity capital?” the percentage of respondents who said they always or almost always
used the CAPM ranged from 34% to 73% across the five countries. However, this is not the
only evaluation technique used in estimating the cost of equity. Another alternative answer
was “using the CAPM but including some extra risk factors.” The proportion of
respondents who stated that they always or almost always incorporated extra risk factors
into the CAPM ranged from 15% to 34%, with an aggregate percentage of 28%. As SFG
report states that if corporate finance practice is used to determine how regulation should
evolve, the benchmark would be to use the factor models (the CAPM with additional risk
factors).
- Factor-based investment
Factor investing has become a widely discussed part of today’s investment canon.
Investment researchers use multifactor risk models to run controlled back tests of future
investment strategies. For this, their needs are similar to those of portfolio managers. They
need to implement strategies optimally on historical data and understand the subsequent
performance of those strategies. Researchers can use back tests to enhance their strategies.
They can also use performance analysis and portfolio risk characterization to improve their
understanding of the bets they are testing. The evidence that factor-based investment is in
vogue nowadays can also be found in the latest special issue (2017) of Journal of Portfolio
Management, two articles (Dimson et al. (2017) and Kim et al. (2017)) focus on factor-
3
SFG Consulting report of the Fama-French model.
6
Introduction
based investing and two articles (Podkaminer (2017), Alford and Rakhlin (2017)) about
smart beta which is a currently popular factor-based investment strategy. For example,
MSCI has created a family of factor indexes (smart beta indexes) that provide access to six
solidly grounded factors—value, low size, low volatility, high yield, quality, and
momentum.
Of course, those mentioned above are not all the applications of factor models in practice.
Other examples like pension plan sponsors can use multifactor risk models to coordinate
their multiple managers. Portfolio risk characterization allows them to understand any gaps
or overlaps among their managers or in their asset allocation mixes. Plan sponsors also use
performance analysis to assess their managers' value added and to check on their managers'
styles.
The extensive research and widely practical use of factor models are two of the motivations
for this dissertation, another one is the Chinese stock market. Significant economic growth
in China over the past few decades has been universally acknowledged, China now has the
second-largest economy in the world, and is the biggest developing country. The Chinese
stock market, established in 1990, is one representative of the emerging markets with
relative short history. Its development along with its immaturity has attracted considerable
attention from researchers, and also has brought into question of whether the asset pricing
models such as FF model are also applicable to its domestic markets.
Therefore, both the popular of factor models and the attractive emerging stock market
prompt the original intention in the study of factor models (risk factors) on Chinese stock
market. One may say that this again another study of factor models on the basis of FF3F
Model, but what we want to emphasize is that FF3F Model has become the cornerstone of
factor studies, the research on factor models have been a hot research topic for decades, and
it will continue to be a key topic of asset pricing field in the future.
Objective and Structure of Dissertation
The work presented in this dissertation contains three chapters, mainly explore the risk
factors on Chinese stock market on the basis of FF3F Model, and the studies of each
chapter are closely linked while relatively independent as well. The success of FF3F Model
on U.S. stock market impels the studies of the applicability in other developed or emerging
markets all around the world. China has attracted more and more researchers’ attention, the
applicability of FF3F Model on Chinese stock market has also been a hot topic for years.
This dissertation begins with re-testing the FF3F Model using data of Chinese A-share
7
Introduction
(CNAS) stock market in order to answer the question that whether FF3F Model is
applicable on CNAS stock market, and the following work is then unfolded on the basis of
FF3F Model.
Chapter 1 re-investigates the FF3F Model and explore the latest FF5F Model on CNAS
stock market, aims to examine the applicability of both models in China. Furthermore, we
compare the regression results of both models to test whether FF5F performs better than
FF3F Model in capturing the variation of average stock returns as they are proved so in U.S.
market. It is worth noting that we take into account several special features of Chinese
stock market that are non-negligible in order to benefit a more reliable research results.
To implement the investigation, we construct all five factors – excess market return, size
factor SMB, value factor HML, profitability factor RMW and investment factor CMA
using data of CNAS stock market, Fama-MacBeth (1973) two-stage approach is applied for
the time-series regression (TSR) and cross-sectional regression (CSR). Since the Fama-
MacBeth (FM hereafter) two-stage approach caused the classical errors-in-variables
problem, the correction procedure proposed by Shanken (1992) is applied.
Though the great achievement of FF factor models, it is also one of the most controversial
asset pricing model, mainly for its purely empirical-based considerations and lack of
theoretical underpinnings. The success of FF3F Model on CNAS stock market stimulates
the chasing down for economic foundation of FF factors, especially the SMB factor. In
chapter 2, we proceed with the research in line with one of the dominant explanations,
which is based on time-varying investment opportunities in the context of ICAPM. We
examine whether FF factors proxy for the innovations of selected state variables that
describe the future investment opportunities on CNAS stock market, and we choose four
economic variables in accordance with Petkova (2006): aggregate dividend yield, one-
month T-bill rate, term spread and default spread, which in order to model two aspects of
investment opportunity set – the yield curve and the conditional distribution of asset returns.
To extract the innovation terms, the vector auto-regression (VAR) process has been utilized;
and as Campbell (1996) emphasizes that “it is hard to interpret estimation results for a VAR
factor model unless the factors are orthogonalized and scaled in some way”, we
orthogonalize innovations of state variables to the excess market return, which is the first
element of the vector in VAR process. The Fama-MacBeth (1973) two-stage approach is
then implemented for both TSR and CSR for the research.
In chapter 3, we choose distress risk factor as the augmented factor of the original FF3F
Model among plenty risk factors that have been discovered. Since firm’s distress risk is an
important indicator of a firm’s performance and is also one of the most concerned
characteristics by investors and firms. What’s more, the financial distress risk has been
8
Introduction
proved closely related to stock returns and the viewpoint of value effect is regarded as the
distress risk effect are proposed by time-series literature.
The main objective is to test whether distress risk factor FF factors are proxies for distress
risk on CNAS stock market during the period July 2005 to May 2015, if not, whether
augmented four-factor model better explains expected average stock returns than FF3F
Model. Especially, we apply both accounting-based (Ohlson’s O-score) and market-based
(Vassalou and Xing’s DLI) models to measure the financial distress in order to identify
whether different methods of predicting distress risk matter to the empirical results. In
addition, the relationship between stock returns and distress risk, and whether the size and
value effects are related to distress risk have also been examined on CNAS stock market.
Contributions and Discussion
Though it seems that we are one of the countless studies of testing FF3F Model on Chinese
stock market initially in Chapter 1, actually we consider several special features of Chinese
stock market and examine not only the time-series but also cross-section validation of FF3F
Model. Furthermore, we examine the performance of the latest FF5F Model using data of
CNAS stock market, while the comparisons between the performance of both models on
CNAS stock market, and the performance of same models between Chinese and U.S. stock
market are also implemented.
Despite the extensive evidence that confirm the capability of FF factors in capturing
variations of expected average stock returns on Chinese stock market, however, to the best
of our knowledge, few studies have authentically focused on the economic underpinning
behind FF factors in China. We are the advance ones who explore one of the most popular
explanations which is based on time-varying investment opportunities in the context of
ICAPM, examine whether FF factors proxy for innovations of selected economic variables
on Chinese stock market.
Moreover, developing an early warning system for prediction of firms’ financial distress
has been a hot topic over the years in China; the research on Chinese market are mainly
focused on testing the predictive accuracy of commonly used models in estimating distress
risk and measuring default risk of Chinese companies without studying the relationship
between default risk and stock returns. We are also the rare ones who propose a mimicking
distress factor and investigate the distress risk factor in the frame of FF3F Model.
Overall, our research has new implications in practice (such as for the factor-based
investment and assessing the performance of portfolios) on Chinese stock market. The
9
Introduction
Chinese stock market developed along with extraordinary economic growth and overall
institutional reforms so that it might be much different from the mature stock markets such
as U.S. stock market. Therefore, for further research, it might be more appropriate to
construct risk factors that feature Chinese own characteristics and introduce which to the
factor model, thus help researchers to understand the real economic meaning of FF factors
on Chinese stock market.
10
1 Fama-French Five-Factor Model vs. Fama-
French Three-Factor Model in China
1.1 Reviews of the main asset pricing theories and models ........................................ 12
1.2 Retrospective of empirical work of Fama-French factor models .......................... 16
1.3 Introduction of Chinese stock market .................................................................... 31
1.4 Empirical Analysis and Results of FF3F Model .................................................... 38
1.5 Construction of profitability and investment factors and empirical results of Fama-
French Five-Factor Model ................................................................................................ 53
1.6 Conclusions ............................................................................................................ 64
This chapter re-examines the applicability of Fama-French Three-

Factor Model by constructing Fama-French factors on Chinese A-
share stock market, considering several special features of Chinese
stock market. In addition, this chapter also investigates the latest
FF5F Model using data of Chinese stock market by apply Fama-
MacBeth two-stage approach. Empirical results answer the
questions of whether Fama-French Three-Factor Model explain
time-series and cross-sectional variation of average excess stock
returns over the sample period; whether the profitability and
investment factors have extra explanatory power and whether
Fama-French Five-Factor Model performs better than Fama-
French Three-Factor Model on Chinese A-share stock market.
Furthermore, the comparisons of the empirical results between
Chinese stock market and U.S stock market are conducted as well.
11
Chapter 1 Fama-French Five-Factor Model vs. Fama-French Three-Factor Model in China
1.1 Reviews of the main asset pricing theories and models
1.1.1 The evolution of the asset pricing models
The history of the theory of the financial asset pricing can be traced back to Bernoulli's
famous St Petersburg paper of 1738, since then, researchers have been focusing on this
subject for centuries. In the early life of financial fields, describing the market environment
and valuing individual securities are the emphasis, which are transformed with the
appearance of Markowitz (1952) paper on portfolio selection. After the birth of Portfolio
Selection theory, the discussions of pricing on the financial asset can also be divided into
two different periods. Before the 1980s, researchers mainly focus on pure theory, applying
Markowitz’s optimal selection theory into their research. But after the 80s, as Sharpe
(1964), Lintner (1965) and Black (1972)’s CAPM becoming a research paradigm, the focal
point has transferred to two directions: extending the original CAPM and the empirical
research of the model.
The initial extension involved extending the single-period model into a multi-period
framework, which turns out to be one of the major developments of modern finance.
Merton (1973a) proposes an Intertemporal CAPM (ICAPM) and shows that the classical
CAPM would not in general hold in continuous time periods. Breeden (1979)’s
Consumption CAPM (CCAPM) links the consumption and stock returns and the model
relies on the aggregate consumption in order to understand and predict future asset prices
instead of the market portfolio's return in the traditional CAPM.
Meanwhile, Jensen et al. (1972) performed the first strict tests of the original CAPM by
constructing ‘’two-pass’’ methodology. While the researchers were trying by all means to
test the CAPM, Roll (1977) shows that the CAPM never could be tested unless the market
portfolio were known with certainty. Even many researchers have tried to tackle the Roll’s
critique, unfortunately, it turns out that CAPM seems not to hold and there are amounts of
evidence that other risk factors also affect the stock returns. The factors will be listed in
section 1.2 below.
Around the same time Ross (1976) develops the Arbitrage Pricing Theory (APT) as an
alternative model that could potentially cover the CAPM’s short falls, while still retaining
the underlying message of the latter. The APT starts by assuming that there are n factors
that cause asset returns to systematically deviate from their expected values, however, the
theory does not specify how large the number n is, nor does it identify the factors.
Empirically, the APT has been investigated by using either factor analytic methods (Roll
and Ross, 1980), to estimate multiple measures of systematic risk, or pre-specified
macroeconomic factory (Chen et al., 1986), in which the studies look at pricing relative to a
12
set of observable macroeconomic variables, or factors, selected primarily based on

economic intuition.
1.1.2 Reviews of Fama-French Three-Factor Model
The facts that there are other risk factors can also explain the variation of stock returns have
been verified by a huge number of research. The factors which are already been verified
include the earnings/price ratio (Basu, 1977), company size (Banz, 1981), book-to-market
equity (Fama and French, 1992) and a variety of other systematic influences on asset price
(Dimson and Mussavian, 1998). An asset pricing model who can better explain the risk and
the stock returns is demanded in the financial market and many researchers are devoted to
do this work. Among which, Fama and French (1993) [hereafter FF] deduce a three-factor
model as an alternative way to predict the stock returns or an extension of the original
CAPM, and their model not only reveals the primary factors that drive stock return but also
provides a strategy for using those factors in the portfolio for a potentially higher expected
long-term return, which makes it an extraordinary model and also represent another
important progress in the history of asset pricing development.
In fact, Fama and French (1992) studied the joint roles of market beta, size, leverage,
Earnings/Price (E/P) ratio and book-to-market equity (B/M) ratio in the cross-section of
average stock returns for NYSE, Amex and NASDAQ stocks over the period 1963-1990. In
that study, the authors find that beta has almost no explanatory power. On the other hand,
when used alone, size, E/P, leverage and B/M ratio have significant explanatory power in
explaining the cross-section of average returns. When used jointly, however, size and B/M
equity ratio are significant and they seem to absorb the effects of leverage and E/P in
explaining the cross-section average stock returns. Fama and French (1992) therefore
argued that if stocks are priced rationally, risks must be multidimensional.
Fama and French (1993)’s analysis was extended to both stocks and bonds. Monthly
returns on stocks and bonds were regressed on five factors: returns on a market portfolio, a
portfolio for size and a portfolio for the book-to-market equity effect, a term premium and a
default premium. For stocks, the first three factors were found to be significant and for
bonds, the last two factors. As a result, Fama and French (1993) construct a three-factor
asset pricing model for stocks that includes the conventional market beta and two additional
risk factors related to size and book to market equity. They find that this expanded model
captures much of the cross section of average returns amongst US stocks.
Compared to the original CAPM, FF3F Model shows that the expected return on a portfolio
in excess of the risk-free rate is explained by the sensitivity of its return to three factors:
13
first, the excess return of the market portfolio; second, the difference between the returns of
the small size portfolio and those of the big size portfolio (SMB) and third, the difference
between the returns of the high book-to-market portfolio and those of the low book-to-
market portfolio (HML). The model is as followed:
Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ei ,t (1.1)
Where:
Ri ,t is the value-weighted return on portfolio i in period t;
R f is the risk-free rate;
RM ,t  R f is the difference between the return of market portfolio (market beta) and the
return of risk-free rate; SMB is the size factor (small minus big) and HML is the value
factor (high minus low);
bi is the coefficient for the excess return of the market portfolio;
si is the coefficient for the excess return of portfolios with small equity class over
portfolios of big equity class;
hi is the coefficient loading for the excess average returns of portfolios with high book-to-
market equity class over those with low book-to-market equity class;
ei ,t is the error term for portfolio i at time t.
1.1.3 Reviews of Fama-French Five-Factor Model
Motivated by the valuation theory and recent empirical findings on the strong profitability
and investment effects in asset returns4. Fama and French (2015a) propose a five-factor
model contains the market factor and factors related to size, book-to-market equity ratio,
profitability and investment, which performs better than the three-factor model of Fama and
French (1993):
4
Recently, Novy-Marx (2013) identifies a proxy today that predicts expected earnings tomorrow -
the profitability factor, which is strongly related to average stock return, and the investment factor
was documented by Aharoni et al. (2013), see also Titman et al. (2004), although it has a high
correlation with the value and profitability factors, the investment effect is perhaps half as strong,
but it is still reliable and significant.
14
Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ri RMW  ciCMA  ei ,t (1.2)
From equation (1.2), it is obvious that the five-factor model has two more factors than the
three-factor model, RMW and CMA. RMW is the factor related to firm’s profitability
which is the difference between the returns on portfolios of robust (high) profitability and
weak (low) profitability firms. CMA is the one related to investment, which is the
difference between the
returns of conservative (low) investment portfolios and aggressive (high) investment

portfolios.
In this paper, Fama and French suggest that the theoretical starting point is the “Dividend
Discount Model”:

mt   E( d t  ) / ( 1  r ) (1.3)
1
where mt is the share price at time t, E( d t  ) is the expected dividend per share for the
period t   , and r is (approximately) the long-term average expected stock return or, more
precisely, the internal rate of return on expected dividends. This model states that the value
of a stock today will be the sum of the discounted present value of all its future dividends.
With a little bit manipulation, the dividend per share d t  is the difference between Yt  , the
equity earnings for period t+τ, and dBt   Bt   Bt 1 , which is the change in book equity.
Then the dividend discount model (equation (1.3)) becomes:

M t   E( Yt   dBt  ) / ( 1  r ) (1.4)
1
Divided by book equity at time t gives,


Mt 
E(Yt   dBt  ) / ( 1  r )
 1 (1.5)
Bt Bt
Equation (1.5) implies three statements about expected stock returns.
- Firstly, fix everything except the expected stock return r and the current value of
the stock M t , a lower market value M t , or equivalent to a higher B/M ratio implies
a higher expected stock return.
15
- Next, fix everything except the expected earnings Yt  and expected stock returns r ,
more profitable companies which with higher expected earnings have higher
expected returns.
- Finally, controlling for the expected growth dBt  (investment) and expected stock
returns while fixing other elements, firms with higher expected growth in book
equity implies a lower expected return.
The dividend discount model and its transformation indicate the relationship between the
variables and average asset returns. FF point out that the nature of equation (1.3) and
equation (1.5) is the reason why they choose profitability and investment factors to
augment the model.
The construction of profitability factor and investment factor and the portfolios are
demonstrated in the following section.
1.2 Retrospective of empirical work of Fama-French factor models
1.2.1 Overviews of the data and methodologies of FF factor models
1.2.1.1 Construction of Fama-French three factors and portfolios
To do the regressions, Fama and French use stocks in the intersection of American stock
markets (1963-1990), NYSE, AMEX, and NASD to construct the portfolios on size and
B/M ratio, among which, financial firms and firms with negative B/M equities are
eliminated from the sample. To obtain the size and value portfolios, they use a firm’s
market capitalization at June of year t to measure its size and at the end of December of
year t-1 to compute book-to-market equity ratio. So in June of each year t, the stocks are
sorted into two size groups: small firms (S) and big firms (B), according to their market
value. They also break stocks into three B/M equity groups at each December of year t-1:
low B/M equity ratio (L), medium B/M equity ratio (M) and high B/M equity ratio (H)
firms, according to the breakpoint 30% and 70% of values of B/M equity for all the stocks.
After these steps, they have two size groups and three B/M equity groups at each year t.
The intersections of these groups are constructed into six portfolios: small low (SL), small
medium (SM), small high (SH), big low (BL), big medium (BM), and big high (BH)
portfolios. For instance, The SL portfolio contains stocks in the small size (S) group which
16
meanwhile have low B/M equity ratio (L); the BH portfolio contains stocks in the big size
(B) group that also in the high B/M equity group (H). The value-weighted monthly returns
are calculated from July of year t to June of year t+1, during which the portfolios remain
the same, and the portfolios are reconstructed in July of year t+1 (B/M ratio should exist at
December of year t-1).
Finally, the Fama-French three factors obtain as follows. The market factor is the excess
market return which is computed as the difference between the value-weighted returns of
all A-shares and the risk-free rate. The SMB factor is then the difference between the
simple average of the monthly returns of the three small-size portfolios (SL, SM, and SH)
and the simple average of monthly returns of the three big-size portfolios (BL, BM, and
BH). Similarly, HML is equal to the simple average monthly return of the two portfolios
with high book-to-market equity (SH and BH) minus which of the two portfolios with low
book-to-market equity (SL and BL).
1
SMB  ( Small Low+Small Medium+Small High)
3 (1.6)
1
 ( Big Low+Big Medium+Big High )
3
1 1
HML  ( Small High+Big High )  ( Small Low+Big Low ) (1.7)
2 2
1.2.1.2 Construction of profitability factor and investment factor
Similar to FF three factors that are constructed using the 6 value-weighted portfolios
formed on size and book-to-market equity. The Fama-French 5 factors (2x3) are
constructed using the 6 value-weight portfolios formed on size and book-to-market (Size-
B/M portfolios), the 6 value-weight portfolios formed on size and operating profitability
(Size-OP portfolios), and the 6 value-weight portfolios formed on size and investment
(Size-Inv portfolios). The Size-OP portfolios and Size-Inv portfolios are formed in the
same way as the Size-B/M portfolios, except the second sort variable is operating
profitability or investment.
The operating profitability (OP) for June of year t is calculated as annual revenues minus
cost of goods sold, interest expense, and selling, general, and administrative expenses
divided by book equity for the last fiscal year end in t-1. The Investment portfolios are
formed on the change in total assets from the fiscal year ending in year t-2 to the fiscal year
ending in t-1, divided by t-2 total assets at the end of each June using NYSE breakpoints.
To be more clear:
17
OP  ( Gross Profitability  Interest Expense 

(1.8)
Selling, General and Administrative Expenses ) / ( Book Equity )t 1
Inv  (Total Asset )t 1  (Total Asset )t 2  / (Total Asset )t 1 (1.9)
Where,
OP represents the operating profitability;
Gross Profitability equals annual revenue minus the cost of goods sold;
Book Equity is book value of equity;
Inv represents the investment opportunities;
( Total Asset )t 1 is the total value of assets in year t-1;
( Total Asset )t  2 is the total value of assets in year t-2.
The size breakpoint for year t is the median NYSE market equity at the end of June of year
t. The construction of portfolios on OP and investment are similar with that of portfolios on
book-to-market equity. At the end of each June, the firms are sorted into three OP
portfolios based on the breakpoints of the 30th and 70th NYSE percentiles, and the three
investment portfolios are formed in the same way using NYSE breakpoints-30th and 70th
NYSE percentiles.
Then at the end of each June, the intersections of two portfolios formed on size - small (S)
and big (B), and three portfolios formed on profitability – weak profitability (W), neutral
profitability (N) and robust profitability (R) are constructed into six “Size-OP” portfolios:
SW, SN, SR, BW, BN and BR 5 . Similarly, the “Size-Inv” portfolios, which are also
constructed at the end of each June, are the intersections of two portfolios formed on size
and three portfolios formed on investment- conservative investment (C), neutral investment
(N) and aggressive investment (A). Thus, the six Size-Inv portfolios are constructed: SC,
SN, SA, BC, BN, and BA6.
In Fama-French five-factor (FF5F) Model, the market factor and value factor remain the
same as in the three-factor model, while the size factor SMB need to be reconstructed with
profitability and investment factors, which is the average return on the nine small stock
portfolios minus the average return on the nine big stock portfolios. The two additional
5
Portfolio SW contains firms with small size and weak profitability, SN contains firms with small
size and neutral profitability, SR contains firms with small size and robust profitability, similarly to
BW, BN and BR, which contains firms with big size and weak profitability, neutral profitability and
robust profitability separately.
6
Portfolio SC contain firms with small size and conservative investment, SN contains firms with
small size and neutral investment, SA contains firms with small size and aggressive investment,
similarly to BC, BN and BA portfolios.
18
factors are directed at capturing the profitability and investment patterns, which are
indicated by RMW and CMA. As shown in equation (1.14) and equation (1.15), RMW is
the difference between returns on portfolios with robust and weak profitability, and CMA is
the difference between returns on portfolios of the stocks of low and high investment firms,
which is called conservative and aggressive, separately. In detail:
1
SMBB / M  ( Small Low+Small Medium+Small High)
3 (1.10)
1
 ( Big Low+Big Medium+Big High )
3
1
SMBOP  ( Small Robust+Small Neutral+Small Weak)
3 (1.11)
1
 ( Big Robust+Big Neutral+Big Weak )
3
1
SMBInv  ( Small Conservative+Small Neutral+Small Aggressive)
3 (1.12)
1
 ( Big Conservative+Big Neutral+Big Aggressive )
3
1
SMB  ( SMBB / M  SMBOP  SMBInv ) (1.13)
3
1 1
RMW  ( Small Robust+Big Robust)  ( Small Weak+Big Weak ) (1.14)
2 2
1
CMA  ( Small Conservative+Big Conservative)
2 (1.15)
1
 ( Small Aggressive+Big Aggressive )
2
1.2.2 Reviews of the empirical tests for Fama-French Three-Factor Model
1.2.2.1 Empirical results of FF on some developed countries’ stock markets
Fama and French (1992) show that two easily measured variables, size and book-to-market
equity, seem to capture the cross-section of average stock returns. Then the article of FF
(1993) extends the FF’s (1992) study by using a time-series regression (TSR) approach,
and the analysis was extended to both stocks and bonds markets. Monthly returns on stocks
19
and bonds were regressed on five factors: returns on a market portfolio, a portfolio for size
and a portfolio for the book-to-market equity effect, a term premium and a default premium.
For stocks, the first three factors were found to be significant. As a result, FF (1993)
construct a three-factor asset pricing model for stocks that composed by the conventional
market beta and two additional risk factors related to size and book to market equity. They
find that this model captures much of the cross section of average returns of U.S. stock
markets.
In addition, they also find that the slopes on SMB for stocks are related to size, in every
B/M ratio quintile of stocks, the slopes on SMB decrease significantly from smaller-size
quintiles to bigger-size quintiles. Similarly, the slopes on HML are related to B/M ratio, in
each size quintile, slopes of HML increase from the lower B/M quintile to the higher B/M
quintile. Researchers have suggested the following possible explanations for the size effect.
Small firms’ stocks are more illiquid and trading in them attract greater transaction costs;
there is also less information available about small firms and therefore the cost of
monitoring a portfolio of small stocks will generally be greater than that of a portfolio of
large firms, and also given that small shares trade less frequently, their beta estimates might
be less reliable. However, all these remain hypothetical explanations for the size effect, as
there is no rigorous theory explaining convincingly why the size effect should be present.
The B/M ratio effect shows that average returns are greater the higher the book-to-market
value ratio and vice versa. It is also referred to as the value premium. The high book value
firms are underpriced by the market and are therefore good buy and hold targets, as their
price will rise later. This anomaly undermines the semi-strong form efficiency of the
market. These two variables explain average return differences across portfolios that cannot
be accounted for by beta.
Fama and French (1995) analyze the characteristics of firms with high B/M ratio and those
with low B/M ratio. They find that firms with high B/M ratio tend to be persistently
distressed and those with low B/M ratio are associated with sustained profitability. They
conclude that the returns to holders of high B/M ratio stocks are therefore a compensation
for holding less profitable and riskier stocks. They show that book-to-market equity and
slopes on HML in the three-factor model proxy for relative distress. Weak firms with
persistently low earnings tend to have high B/M ratio and positive slopes on HML; strong
firms with high earnings have low B/M ratio and negative slopes on HML.
Similarly, Chan and Chen (1991) posit that small and large firms have different risk and
return characteristics. Small firms on the New York Stock Exchange are firms that have not
been doing well, are less efficiently managed and are highly levered. As a result, small
firms tend to be riskier than large firms and that risk is not captured by the market index.
After introducing multiple risk exposures to the market index; a leverage index and a
20
dividend-decrease index to mimic the marginal firms, the size effect loses its explanatory
power. Risk exposures to these indices are as powerful as size in explaining average returns
of size-ranked portfolios.
However, Kothari et al. (1995a) and MacKinlay (1995a) argue that a substantial part of the
premium is due to survivor bias and data snooping. The data source for book equity
contains a disproportionate number of high B/M firms that survive distress, so the average
return for high B/M equity firms is overstated. The data snooping hypothesis posits that
researcher’s fixation to search for variables that are related to average return, will find
variables, but only in the sample used to identify them. But a number of papers have
weakened and even dismissed the survivorship-bias and the data snooping hypothesis. For
instance, Lakonishok et al. (1994a) find a strong positive relation between average returns
and B/M ratio for the largest 20 percent of NYSE-Amex stocks, where survivor bias is not
an issue. Similarly, FF (1993) find that the relation between B/M ratio and the average
return is strong for value-weight portfolios. As value-weight portfolios give most weight to
larger stocks, any survivor bias in these portfolios is trivial. There are also are many studies
using different sample periods on US data and samples in different countries confirming the
existence of the size and book-to-market equity effects. FF (1998) provide additional
valuable out-of-sample evidence. They tested the FF three-factor model in thirteen different
markets over the period 1975 to 1995. They find that 12 of the 13 markets record a
premium of at least 7.68 percent per annum to value stocks (high B/M ratio). Seven
markets show statistically significant B/M betas.
Maroney and Protopapadakis (2002) test the FF three-factor model on seven stock markets
of Australia, Canada, Germany, France, Japan, the UK and the US. The size effect and the
value premium survive for all the countries examined. They conclude that the size and B/M
effects are international in character. Using a Stochastic Discount Factor (SDF) model, and
a variety of macroeconomic and financial variables, do not diminish the explanatory power
of B/M ratio and market value. Their evidence suggests that the B/M and size effects are
not artifacts of the inadequacies of the augmented CAPM as an asset-pricing model or of
omitting macroeconomic and financial variables. The positive relation of returns with B/M
and their negative relation with market value remain strong under a general SDF model.
Faff (2001) uses Australian data over the period 1991 to 1999 to examine the power of the
FF3F Model. He finds strong support for the FF3F Model, but a significant negative rather
than the expected positive, premium, to small size stocks. Faff (2001) concludes that his
results appear to be consistent with other recent evidence of a reversal of the size effect.
Gaunt (2004) studies the FF3F Modl in the Australian setting and provides further out of
sample (non-US) tests of the model. The study covers the period 1991 to 2000 of firms
listed on the Australian Stock Exchange. He finds that beta risk tends to be greater for
21
smaller companies and those with lower BM ratios. Contrary to FF, the betas are on
average significantly less than one. There is also evidence of the value effect increasing
monotonically from the lowest to the highest book-to-market equity portfolios. There is a
monotonic increase of loadings on the SMB factor as well when moving from the largest to
the smallest portfolios. They find large and positive intercepts for the small portfolios. The
explained variation as measured by the adjusted 𝑅 2 is also much higher compared to the
CAPM. The author concludes that the three-factor model provides a better explanation of
observed Australian stock returns than the CAPM.
Numerous studies examined various global markets following research of FF. Studies on
the UK market by Chan and Chui (1996) fail to find a significant beta-return relationship.
Subsequent studies by Hung et al. (2004) and Morelli (2007), again examining the UK
market all provide further evidence against an unconditional relationship between beta and
returns. Morelli (2007) find a significant relationship between beta and returns in the
presence of size and book-to-market equity, but the author documents that size is not a
significant risk variable, whereas B/M ratio is a significant determinant of security return
on UK market. Studies of other European markets including, Lilti and Montagner (1998) in
the French market, Isakov (1999) in the Swiss market, and Elsas et al. (2003) in the
German market, all find beta to be a significant risk measurement
1.2.2.2 Summary of empirical results in emerging Asian markets
Most Asian stock markets are emerging market which draws researchers’ attentions, and
it’s interesting and meaningful to study the FF3F Model which is well performed in U.S. on
the emerging Asian markets. Researches are done which concerning Fama-French factors
and stock market anomalies on the emerging stock market in Asia: such as India, Thailand,
Indonesia, Malaysia, Singapore, Korea, Vietnam and Hong Kong.
Early in 1995, Claessens et al. (1995) study the cross-section of stock returns from twenty
emerging markets, among which, several Asian markets (Indonesia, Korea, Malaysia,
Philippines, Thailand) are included. They find that besides market risk, firm size, E/P ratio
and turn over factors are significant in explaining a cross-section of stock returns. Among
those Asian countries, the market beta has no explanatory power on most of the stock
markets, only India, Malaysia and Thailand stock market have size effect. Chui and Wei
(1998) conduct empirical tests on the robustness of the multifactor model in the Asian
region. They examine the relationship between expected stock returns and market beta, size
and B/M equity in five Asian emerging stock markets: Hong Kong, Korea, Malaysia,
Taiwan, and Thailand. Their results suggest a weak relationship between average stock
22
returns and the market beta for all five markets. But the book-to-market equity can explain
the cross-sectional variation of expected returns in Hong Kong, Korea, and Malaysia, while
the size effect is significant in all five markets except Taiwan. Drew (2003) compares the
explanatory power of FF3F Model in Hong Kong, Korea, Malaysia and Philippines, and the
author documents that size and value effects exist for all four markets but the multifactor
model of FF provides a parsimonious description of the cross-section of returns for these
Asian markets over the 1990s. Furthermore, Shum and Tang (2005) apply FF3F Model on
three Asian emerging markets: HK, Singapore and Taiwan over the period July 1986 to
December 1998, and they conclude that the model largely explains the variations in average
returns when using the contemporaneous market factor, but the impact of size effect and
value effect is very limited and insignificant in most cases. Sehgal et al. (2014) test the
equity market anomalies for six emerging market including China, India, Indonesia, South
Korea and South Africa. They find that there is size effect in India and both size and value
effect in South Korea.
Connor and Sehgal (2001) find a pervasive influence of FF3F model on random stock
returns in India from 1989 to 1998, however, the market factor alone cannot explain the
cross-sectional stock returns. Sehgal and Balakrishnan (2013) re-examine the robustness of
CAPM and FF3F model using data from 1996 to 2010, and results turn out that FF3F
model does a better job in explaining the returns on most portfolios constructed based on
firm size and B/M equity ratio. Ranjan Dash and Mahakud (2013) propose a multifactor
model which include liquidity, market leverage, and momentum along with size and value
factors to investigate the firm-specific anomaly effect and to identify market anomalies that
account for the cross-sectional regularity in the Indian stock market. Though the anomaly
effect is weak, their five-factor model is able to capture the B/M equity, liquidity, and
medium-term momentum effect, and size, market leverage, and short-run momentum effect
are found to be persistent in the Indian stock market. Based on the data of period Jun 2001
to Jun 2006, Jun 2004 to Jun 2009, Oct 1998 to Sep 2013, Jan1997 to Jun 2012 and Jan
1997 to Aug 2014, separately, Bahl (2006), Taneja (2010), Das (2015), Balakrishnan (2014)
and Balakrishnan (2016)’s research continues to prove the existent of the size and value
effect in the Indian stock market.
In other Asian emerging markets, Ferdian et al. (2011) find that market beta alone is not
sufficient to describe the variation in average equity returns, but there exist size and value
premium for Indonesian Shariah Stocks over the period September 2007 to September 2009.
However, Sudiyatno and Irsad (2013) find no prove of size and value effects but a
significant positive relationship between the risk premium and stock returns over the period
2007 to 2009 in Indonesia stock exchange. Amanda and Husodo (2014) explore FF3F
Model along with liquidity in Indonesia from 2003 to 2013, and they conclude that there
23
are market risk premium, size premium, value premium and liquidity risk premium which
can explain the excess return in Indonesia.
Lau et al. (2002) examine the relationship between stock returns and beta, size, the
earnings-to-price ratio, the cash flow-to-price ratio, the book-to-market equity ratio, and
sales growth. And they proved the existence of anomalies in Singapore and Malaysia for
the period June 1988 to December 1996. Drew and Veeraraghavan (2002) present evidence
of the size and value premium for the case of the Malaysia. They report that the factors
identified by FF explain the variation in stock returns in Malaysia and are not sample
specific. The analysis was restricted to firms with available returns data from December
1992 to December 1999. Their findings clearly document evidence of a size and B/M
equity effect that small and high B/M equity stocks generate higher returns than big and
low B/M stocks in Malaysia. The results also show that the explanatory power of the
variables is powerful throughout the sample period and not solely in January.
Lam (2002) investigates the relation between stock returns and size, leverage, B/M ratio,
and earnings–price (E/P) ratio in Hong Kong stock market using FF approach. And he
proves that market beta is unable to explain the average monthly returns on stocks listed in
Hong Kong Stock Exchange over the period July 1984 to June 1997. But size, book-to-
market equity, and E/P ratio seem able to capture the cross-sectional variation in average
monthly returns over the period. Other two variables book and market leverages, are also
able to capture the cross-sectional variation in average monthly returns. But their effects
seem to be dominated by size, book-to-market equity, and E/P ratio, and considered to be
redundant in explaining average returns when size, book-to-market equity, and E/P ratios
are also considered. Ho et al. (2006) examine whether market beta, size and B/M ratio are
priced under different market conditions on Hong Kong stock market, such as the up or
down market. It is found that, for the whole market, size and B/M ratio were priced but not
market beta. However, separate the whole market into up and down markets, there exists a
significant systematic relation between market beta and stock returns. But size effect is
insignificant in up markets, and B/M effect is negligible in down markets
Moreover, Homsud et al. (2009) prove the efficiency of FF3F model in The Stock
Exchange of Thailand over the period July 2002 to May 2007 and they conclude that FF3F
model can explain risk in stock returns better than the traditional CAPM. Phong and Hoang
(2012) assess the application of FF3F Model in Vietnam's stock market from Jan 2007 to
Dec 2011, and the results show that FF3F Model explaining average stock returns superior
to CAPM.
In general, there are pervasive size and value effects while the market beta seems not able
to explain cross-sectional stock returns on most Asian emerging markets. Thus, the FF3F
24
Model performs better than CAPM in explaining the anomalies of stock returns, especially
size and value anomalies according to the researches mentioned above.
1.2.2.3 Empirical tests in Chinese stock market
As the biggest emerging market in Asian, Chinese stock market has attracted considerable
attention, the cross-sectional relationship between firm-specific characteristics, such as firm
size and B/M equity, and stock returns is also an intriguing issue for researchers. Previous
studies identify several important factors in explaining stock returns in various stock
markets around the world. There is limited literature that investigate the CNAS market in
this respect until recent years (after 2010) and more researchers consider the special
features of Chinese stock market, which we will precise in next section. This part will
review the studies that apply the FF3F Model to Chinese mainland stock market returns.
Drew et al. (2003) use data of Shanghai Stock Exchange for the period 1993 to 2000, and
test the multifactor approach to asset pricing in China considering the non-tradable and
tradable market value. Their analysis suggest that market beta alone is not sufficient to
explain the variation in the average stock returns, and small and growth firms generate
superior returns than big and value firms, but value effect is not as pervasive as was found
for the US portfolios and other international markets. In addition, the authors also conclude
that there is no proof of seasonal effects, such as the January and/or Chinese New Year
effect, which can determine the size and value effects. However, they only run the one-
stage time series regressions using mimic portfolios based on size and book-to-market
equity, no cross-sectional pricing analysis was conducted.
Wang and Xu (2004) investigate FF3F Model on CNAS stock market including both the
Shanghai and Shenzhen A-share stock markets for the sample period July 1996 to June
2002. Firm size is proved to explain the cross-sectional differences in average stock returns,
but contrary to the findings on U.S. market, their results showed the B/M ratio had no effect
on the Chinese stock markets.
Both Eun and Huang (2007) and Wang and Di Iorio (2007) explore the cross-sectional
relationship of average stock returns and several Chinese firms-specific characteristics in
the CNAS stock market and both find that market beta lacks the explanatory power even
when the beta effect is examined respectively. Wang and Di Iorio (2007) propose that the
absence of a market beta effect may be attributed to some specific market characteristics in
the A-share market, such as government intervention, irrational behavior of individual
investors, and the prohibition of short sales (reference to Kang et al. (2002), Hu (1999) and
Wang (2004)). However, their results suggest the existence of value effects in addition to
25
size effect over their research periods, which is contrary to the findings of Wang and Xu
(2004).
Rutledge et al. (2008) investigate whether there is a size effect in case of bull and bear
markets in Chinse stock market over a six-year period from 1998 to 2003. And they find
small firms have significantly greater positive excess returns than large firms during the
bull market, however, small firms have significantly greater negative returns (using total
market value), or no significant difference in returns (using float market value) during the
bear market period.
Wong et al. (2006) and Morelli (2012) both focus on the A-share stock market of Shanghai
Stock Exchange in China. The former explores the cross-sectional stock return behavior
and estimate the beta, size and value effects over the period 1993 to 2002, results show that
smaller firms and value stocks perform better. Systematic risk is negatively significant in
down markets. The latter explores the cross-sectional relationship between security returns
and beta, size and book-to-market equity, and there is no proof of an unconditional
relationship between market beta and returns. However, a conditional relationship is found
when the data is split into up and down markets, the relationship holds even in the presence
of size and book-to-market equity. Both size and book-to-market equity are found to
display both an unconditional and conditional relationship with stock returns.
Researches are applied on Chinese stock market during recent years, especially after 2010.
Xuanjuan Chen et al. (2010) take 18 firm-specific variables which are demonstrated to
predict cross-sectional stock returns in U.S and examine their relationship with stock
returns in Chinese stock market over the period 1995 to 2007. But only five variables are
able to predict stock returns (B/M equity, assets growth, and illiquidity, etc.), and not as in
U.S. market, in multivariate regression tests few variables succeed in multivariate
regression tests.
Contradict with most of the existing literature that there exist size and value effect across
the returns of Chinese stock market7, Wu (2011) apply FF3F model on CNAS markets:
Shanghai Stock Exchange (SSE) and Shenzhen Stock Exchange (SZSE), separately, June
1992 to April 2009 for SSE and February 1996 to April 2009 for SZSE. The results of
CSRs suggest that market beta has no explanatory power when examined alone or with
other factors, and there is evidence of value effect on SSE but no size effect was found.
Finally, the author has the conclusion that the model works better in SSE than in SZSE.
Contradictorily, Chen et al. (2015) find a strong size effect but no value effect cross returns
based on the data period July 1997 to December 2013 in Chinese stock market, the results,
which is consistent with Wang and Xu (2004).
7
Eun and Huang (2007), Wang and Di Iorio (2007), Cakici et al. (2015) and Gan et al. (2013).
26
It is noteworthy that more and more researchers take the special features of Chinese stock
market into consideration. For instance, Chen (2004) 8 examines the performance of the
Fama-French three-factor model for CNASs. The author sorts stocks by their tradable
shares’ market value into three size groups using breakpoints at the 30% and 70%
percentiles. The portfolio returns are value-weighted by the tradable shares’ market value.
Mao et al. (2008)9 apply the FF3F Model to study the long-run return performance after
Chinese listed firms completed rights offering. To construct the three factors, they also sort
stocks into two size groups by their tradable shares’ market value. Liao and Shen (2008)10
use FF3F Model to examine stock price reaction to Chinese listed firms’ completion of the
split-share structure reform that was initiated in April 2005. To construct the size factor,
they separate small and large stocks by the median of their tradable shares’ market value,
which is defined as the number of tradable shares at the beginning of each year multiplied
by share price. To construct the value factor, they use the net assets per share divided by
share price as the B/M ratio because of the market segmentation in China. The portfolio
returns are value-weighted by the tradable shares’ market value, which implicitly assumes
that the portfolios include only tradable shares. In order to examine the explanatory power
of the FF3F Model on Chinese bond returns, Liu and Yang (2010)11 sort stocks by their
price-to-book ratio into three groups, and the portfolio returns are value-weighted by the
total market value; to construct the size factor, they sort stocks by their total market value
into two groups. The results show that two factors, SMB and HML, do not contribute
significantly to explaining Chinese bond returns.
Zhang and Xu (2013) provide an empirical evidence of to what extent the three factors
explain the variation in Chinese stock returns and identify some pitfalls that arise in the
application of the three-factor model to Chinese stock returns. They summarize three
special features of Chinese stock market that potentially affect the three factors
considerably which also have the influence to the explanatory power of FF three-factor
model. In addition to the tradable and non-tradable shares and market segmentation in
Chinese stock market, they also take into account that whether Small Medium Enterprise
8
Chen, Zhanhui. (2004), ‘Cross-sectional Variations and Three Factors Asset Pricing Model:
Empirical Evidence from China A-Share Market’, Chinese Journal of Management Science 6: 13-
18 (in Chinese).
9
Mao, X-Y., Chen, M-G. and Yang, Y-H. (2008), ‘Long-run Return Performance following Listed
Rights Issue: Based on the Improved Three-factor Model’, Journal of Financial Research 5: 114-129
(in Chinese).
10
Liao, L. and Shen, H-B. (2008), ‘Fama-French Three Factors Model and the Effect of the Split-
share Structure Reform’, The Journal of Quantitative and Technical Economics 9: 117-125 (in
Chinese)
11
Liu, G-M. and Yang, C. (2010), ‘Application of Fama-French Multi-Factor Model in China’s
Bond Market during Recent Financial Crisis’, Journal of Zhejiang University (Science Edition) 4:
396-400 (in Chinese).
27
Board (SME) and the Growth Enterprise Board (GEB) listed firms should be excluded in
determining the breakpoints for the size factor. They experiment with different ways of
constructing the three factors in order to evaluate the effect of these special features in
China, the results turn out to be that the formation of the three factors can have a big impact
in empirical studies applying the FF3F Model to Chinese stock market. Overall, FF3F
Model can explain more than 93% of the variation in the portfolio returns on CNASs for
the period of 1996 to 2013, and it does not affect the explanatory power of FF3F model
whether or not the SME and GEB stocks are included to determine the portfolio
breakpoints. The explanatory power of the three-factor model is higher when the market
portfolio includes only tradable shares than when the market portfolio includes both non-
tradable and tradable shares, and when the book-to-price ratio (B/P) are used instead of the
book-to-market ratio.
Though there is less research in Chinese stock market comparing to U.S. and other
developed stock markets, there is time-series literature provide evidence that size or value
(or both) effect(s) on Chinese stock market but the market beta seems not have explanatory
power in explaining the cross-sectional stock returns in China. What’s more, considering
the special features of Chinese stock market seems to be the right direction to apply the
FF3F Model to Chinese stock market as the literature documented, since the explanatory
power of FF3F Model has clearly improved.
1.2.3 Profitability and investment factors and Fama-French Five-Factor Model
Cooper and Maio (2016) state that “The investment anomaly can be broadly classified as a
pattern in which stocks of firms that invest more exhibit lower average returns than the
stocks of firms that invest less”, and “the investment anomaly can be broadly classified as a
pattern in which stocks of firms that invest more exhibit lower average returns than the
stocks of firms that invest less”.
Sloan (1996) is the first who document that accruals are negatively related to future
profitability and that higher accruals predict lower stock returns. Following, an extensive
literature initiated by Sloan (1996), such as Chan et al. (2006) also indicate that accruals are
reliably and negatively related to future stock returns (See also Xie (2001), Richardson et al.
(2005), and Richardson et al. (2006)). Novy-Marx (2013) uncovers a positive relationship
between profitable firms and expected returns that profitable firms earn significantly higher
average returns than unprofitable firms. Haugen and Baker (1996b) and Cohen et al. (2002)
find that controlling for B/M, average returns are positively related to profitability. Fairfield
et al. (2003) find that the well-documented accrual anomaly extends to growth in long-term
28
net operating assets, thus the accrual anomaly documented in Sloan (1996) is a subset of a
larger anomaly with respect to a general market mispricing of growth in net operating
assets.
Working within the confines of a valuation equation, Abarbanell and Bushee (1998),
Frankel and Lee (1998), Dechow et al. (2000), combine analyst’ forecasts of earnings with
assumptions about future investment to estimate expected stock returns. General results
indicate that higher expected net cash flows (expected profitability minus expected
investment) relative to current market value forecast higher stock returns.
Titman et al. (2004) show that firms which increase capital investment tend to have future
negative risk-adjusted returns; and a similar conclusion of a negative relation between
average returns and investment is obtained by Richardson and Sloan (2003). Both
Anderson and GARCIA-FEIJÓO (2006) and Cooper et al. (2008) find that firm-level
investment growth is a robustly significant predictor of the cross-section stock returns,
furthermore, the former propose that the investment anomaly appears to contain
information similar to that of the B/M ratio. Fama and French (2008) investigate the
anomalies which including the asset growth and profitability, and they provide evidence
that higher profitability tends to be associated with abnormally high returns among
profitable firms.
By applying a standard q-theory, Xing (2008a) constructs an investment growth factor,

defined as the difference in returns between low-investment stocks and high-investment
stocks. The author finds that the investment factor contains information similar to the FF
(1993) value factor HML, and can explain the value effect about as well as HML. Similarly,
Lyandres et al. (2008) construct the same investment factor as Xing (2008) and their results
indicate that the investment factor earns a significantly positive average return.
In the paper of Fama and French (2006), they have already studied for the three variables,
B/M ratio, profitability, and investment effects, which are related to expected stock returns
according to dividend discount model and the valuation equation. And they confirm the
implies of valuation theory that high rates of investment are related to low expected returns
when controlling B/M ratio and profitability, while controlling two other variables, high
profitable stocks have higher expected stock returns. Supporting FF valuation perception,
Aharoni et al. (2013) find a positive relation between expected profitability and returns, and
crucially a negative relation between expected investment and returns. They emphasize that
“measuring investment at the firm level rather than per share level is the key to empirically
understanding the simultaneous relation between expected returns, B/M, expected
profitability, and expected investment”.
29
Especially, Hou et al. (2014) examine nearly 80 anomalies in the literature from January
1972 to December 2012 on U.S. market based on q-theory, but about one-half of the
anomalies seems exaggerated their explaining power for average stock returns. They come
to a conclusion that a four-factor model which includes the market factor, size factor,
profitability factor and investment factor explains the cross-sectional average stock returns
to a large extent, and outperforms the FF3F model and Carhart (1997) four-factor models.
Inspired by recently research that give evidence to the remarkable existence of profitability
and investment effects, based on the dividend discount model (equation (1.3)), Fama and
French (2015a) propose a five-factor model contains the market factor and factors related to
size, book-to-market equity ratio, profitability and investment and test the performance of
the five-factor model for the U.S. market using the data from July 1963 to December 2013.
(Data period July 1963-December 2013). They use three sets of factors 12 in order to
examine whether the specifics of factor construction do have an important impact on the
results of the test of asset pricing models. Furthermore, they show GRS statistic of Gibbons
et al. (1989) to test whether the intercepts are indistinguishable from zero in the regressions
of the portfolios’ excess returns on the models’ factor returns so that to distinguish whether
a model can completely capture expected returns.
The results show that the factors from the 2x3, 2x2 and 2x2x2x2 sorts obtain much the
same results in testing of a given model, and although the GRS tests13 indicate that all the
models are incomplete descriptions of expected average returns, the FF5F model
outperforms FF3F model by adding profitability and investment factors. As FF themselves
say, “Despite rejection on the GRS test, the five-factor model performs well: unexplained
average returns for individual portfolios are almost all close to zero”.
Meanwhile, FF find that when the profitability and investment factors are added into FF3F
model, the value factor HML seems to become redundant in explaining average expected
returns. Thus, they draw a conclusion that the value factor HML is a redundant factor for
12
The three sets of factors are: 2x3 sorts on Size and B/M, or Size and OP, or Size and Inv; 2x2
sorts on Size and B/M, or Size and OP, or Size and Inv; and 2x2x2x2 sorts on Size, B/M, OP and
Inv (see details in Fama and French, 2014). 2x3 sorts on Size and B/M is that the size and value
factors are independently sort stocks into two size groups and three B/M groups, and construct the
size factor SMB and value factor HML as of FF3F model; the 2x3 sorts on Size and OP or Size and
Inv are the same as Size and B/M except the sort for B/M groups are replaced by operating
profitability or investment. 2x2 sorts method is similar as 2x3 sorts except that the stocks are all
independently sorted into two groups. In 2x2x2x2 sorts is that the size factor SMB equal weights
high and low B/M, robust and weak OP, and conservative and aggressive Inv portfolio returns.
13
The GRS test examine whether an asset pricing model completely captures expected returns. If
the five-factor model can explain all cross-sectional variation in expected stock returns, then the
intercept will be indistinguishable from zero in a regression of an asset’s excess returns on the
model’s factor returns.
30
describing average returns in FF5F model, which is consistent with the previous findings of
Anderson and GARCIA-FEIJÓO (2006) and Xing (2008a). They explain this outcome is
because “the average HML return is captured by the exposures of HML to other factors
(market factor, SMB, HML and especially RMW and CMA)”.
FF’s (2015a) results suggest that a five-factor model performs better than their FF3F Model.
But the five-factor model fails to capture low average returns on small stocks with high
investment and low profitability. They also show that the model’s performance is not
affected by the way the factors are calculated. With two additional factors, their results also
suggest that the value factor (HML) becomes redundant.
There is not much research on FF Five-Factor model out of America. FF (2015b) proceed
the international tests of FF5F model in North America, Europe, Japan, and Asia-Pacific.
Expected stock returns increase with the B/M ratio and profitability and decrease with
investment for North America, Europe, and Asia Pacific, however, the average stock
returns show little relation to profitability or investment factors. On Brazilian market,
Martinsa and Eid Jr (2015) test the performance of FF5F model during the period January
2002- December 2012 and find that FF5F Model performs better than their previous work
in the three-factor model. The market factor, SMB, and HML capture most of the variation
in average returns in the TSR, however, the two new factors RMW and CMA have shown
less explanatory power. Chiah et al. (2015) investigate the FF5F Model on Australia market,
and they find that the profitability and investment factors have significantly positive
premium. FF5F Model is proved to be able to explain average stock returns better than
FF3F Model in Australia, in contrary to FF (2015a) results, the value factor (HML) remains
its explanatory power in the presence of the investment and profitability factors.
To the best of our knowledge, there is no such a work on Chinese stock market so far. In
next section, we will present the empirical results that we apply FF5F Model on CNAS
stock market.
1.3 Introduction of Chinese stock market
1.3.1 Background of Chinese stock market
In the past 30 years, China experienced extraordinary economic growth and has become an
increasingly important member of the global economy. One of the critical economic
reforms was the introduction and the development of the stock markets. Still young and
31
immature, the Chinese stock markets have grown rapidly and now become the second
largest in terms of market capitalization.
On Chinese stock market, there are three kinds of stocks: A-share stocks, B-share ‘stocks,
and H-share stocks. The ‘A-shares’ do not refer to the ‘class’ of common or preferred
stocks as usual, it refers to shares that are purchased and traded on the SSE and SZSE,
which are two membership institutions governed directly by the China Securities
Regulatory Commission (CSRC). These companies are incorporated in mainland China and
their shares are denominated in the local currency Chinese yuan, or RenMinBi (RMB). For
individual investors, these stocks of the A-share market are strictly off limits to non-
Chinese investors. Meanwhile, some Chinese companies are listed in Shanghai and
Shenzhen, but their shares trade in U.S. dollars. These stocks, known as ‘B-Share’, were
designed to give Chinese companies a way to raise capital from overseas. ‘B-Shares’ also
allow foreign (non-Chinese) investors to invest in this market without the restrictions
associated with ‘A-shares’. ‘H-shares’ are also Chinese companies, but these securities are
traded on the Hong Kong Stock Exchange rather than on the mainland, and they are priced
in Hong Kong dollars.
The Chinese stock market is a young market with relative short history, and it has grown
and expanded rapidly since the establishment of Shanghai stock exchange (SSE) on
November 26th, 1990 - in operation on December 19th of the same year and the other stock
exchange- Shenzhen stock exchange(SZSE) on December 1st, 1990 (opened on July 3rd,
1991). The total listed firms on A-share stock market increase rapidly from 14 of the year
1991 to more than 1000 of the year 2001, and more than 2500 until now (in SSE and in
SZSE). This rapid growth has attracted considerable academic interests; many studies have
examined the ability of FF3F model to predict the stock price movements of Chinese stock.
Figure 1.114 shows the performance of both Shanghai Stock Exchange A-Share (SHASHR)
Index and Shenzhen Stock Exchange A-Share (SZASHR) Index from 1992 to 2015. It
contains two parts, index point (white for SHASHR Index and green for SZASHR) and
volume (grey for SHASHR and rose red for SZASHR). SHASHR Index has averaged
1921.11 index points from 1992 until 2015, reaching the highest 6395.76 index points in
October of 2007 and a record low of 293.75 index points in January of 1992. SZASHR
Index reaches the record high also in 2007 (2015 not included), a record low of 95.26 in
July 1994, and has averaged 681.16 index points from 1991-2015. Both indexes have the
volume increased before 2007 and change unstable after, until 2014 and 2015, both indexes
have a remarkable growth. Figure 1.2 shows the performance of both SHASHR Index and
NYSE Composite (NYA) Index in U.S., in the upper part, the white line is SHASHR Index
14
Source: Bloomberg Finance L.P.
32
Figure 1.1 SHASHR Index and SZASHR Index (1992-2015)
This figure shows the performance of both Shanghai Stock Exchange A-Share (SHASHR)
Index and Shenzhen Stock Exchange A-Share (SZASHR) Index from 1992 to 2015. The
upper part is the index point, white for SHASHR Index and green for SZASHR Index; the
nether part is the volume, grey for SHASHR Index and rose red for SZASHR Index.
Figure 1.2 SHASHR Index and NYSE Composite Index (1992-2015)
This figure shows the performance of both SHASHR Index and NYSE Composite (NYA)
Index from 1992 to 2015. The upper part is the index point, white for SHASHR Index and
green for NYA Index; the nether part is the volume, grey for SHASHR Index and red for
NYA Index.
33
while the green one is NYA Index. Thought the index price of NYA Index is much higher
than that of SHASHR Index, the trends of both indexes are much similar.
Table 1.1 shows the summary of Chinese stock market, the annual total listed firms on the
whole stock market and listed firms respectively on SSE and SZSE from 1990-2014, listed
stocks of A-share stock market, total market capitalization (whole stock market, A-share
Table 1.1 Summary of Chinese stock market 1990-201415
The first column is the year from 1990 to 2014, and total listed firms, listed firms on SSE (denoted
as SSE) and SZSE (denoted as SZSE), listed stocks of A-share (Stocks of A), and total market
capitalization (Total market cap) of whole Chinese stock market, A-share market (A) and B-share
market (B) are represented in the following columns. The last column is the percentage of B-share
value in total market capitalization (B value in total). The unit of total market capitalization is 100
million yuan. The symbol ‘-’ indicate the data that we fail to obtain.
Total Total Total Total
Stocks B value in
Year listed SSE SZSE market market cap market cap
of A total (%)
firms cap of A of B
1990 10 8 2 - - - - -
1991 14 8 6 - - - - -
1992 53 29 24 53 1048 978 70 6.68%
1993 183 106 77 177 3531 3319 212 6.00%
1994 291 171 120 287 3691 3516 175 4.74%
1995 323 188 135 311 3474 3311 164 4.72%
1996 530 293 237 514 9842 9449 394 4.00%
1997 745 383 362 720 17529 17154 375 2.14%
1998 851 438 413 825 19506 19299 206 1.06%
1999 949 484 465 922 26471 26168 304 1.15%
2000 1088 572 516 1060 48091 47456 635 1.32%
2001 1160 646 514 1140 43522 42246 1277 2.93%
2002 1224 715 509 1213 38329 37527 803 2.10%
2003 1287 780 507 1277 42458 41520 937 2.21%
2004 1377 837 540 1363 37056 36309 746 2.01%
2005 1381 834 547 1358 32430 31811 620 1.91%
2006 1434 842 592 1411 89404 88114 1290 1.44%
2007 1550 860 690 1527 327141 324588 2553 0.78%
2008 1625 864 761 1602 121366 120567 800 0.66%
2009 1718 870 848 1696 243939 242127 1812 0.74%
2010 2063 894 1169 2041 265423 263221 2202 0.83%
2011 2342 931 1411 2320 214758 213310 1448 0.67%
2012 2494 954 1540 2472 230358 228775 1582 0.69%
2013 2489 953 1536 2468 239077 237403 1674 0.70%
2014 2613 995 1618 2592 372547 370823 1724 0.46%
15
Source: http://www.stats.gov.cn/
34
stock market and B-share stock market, respectively) and the percent of B-share market
capitalization in total market capitalization from 1992 to 2014. Figure 1.3, Figure 1.4 are
created to visualized Table 1.1. Figure 1.3 shows the total listed firms and that of SSE and
SZSE from 1990 to 2014, listed stocks of A-share stock market are also presented. Figure
1.4 represents the total market capitalization of the whole stock market and also
respectively of A-share stock market and B-share stock market from 1992 to 2014.
From Table 1.1 and the figures 1.3 and 1.4, it is obvious that the Chinese stock market has
undergone a dramatic growth since its establishment in 1990. The total listed firms increase
sharply during 25 years (from 10 total listed firms in 1990 to 1088 in 2000, and 2613 total
listed firms in the year 2014), and total market capitalization from RMB 104.8 billion (year
1992) to 37 254.7 billion (year 2014). The total listed stocks in A-share stock market
increased from 53 (year 1992) to 2592 (year 2014) with a combined market capitalization
of RMB 97.8 billion in 1992 to 37 082.3 billion in 2014.
Figure 1.3 Total listed firms and that of SSE and SZSE, listed stocks of A-share stock market
(1990-2014)
Much of the literature16 that studies Chinese stock market has focused on the segmentation
of the market and mispricing between A shares, denominated in domestic currency, and B
shares, traded in foreign currency. However, this anomaly has been significantly reduced
following the opening of the B market to domestic investors in 200117. The percentages of
B-share market capitalization in total market capitalization show that A-share account for
the vast majority of the total market capitalization (more than 99% after 2006), and B-
shares account for only a very small part of the total market capitalization after 2001 and
even less
16
See Sun and Tong (2000), Chen et al. (2001), Fung et al. (2000) and Fernald and Rogers (2002).
17
Ahlgren et al. (2009)
35
Figure 1.4 Total market capitalization and that of A-share and B-share stock market
(unit:100 million yuan; 1992-2014)
Total market capitalization Total market capitalization of A-share

Total market capitalization of B-share
400000
300000
200000
100000
0
1994
1998
2002
1992
1993
1995
1996
1997
1999
2000
2001
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
than 1% from 2007 (2.93% of year 2001, 0.46% of year 2014). Our research will focus only
on CNAS stock market.
1.3.2 Special features of Chinese stock market
The emerging empirical literature suggests that Chinese market has some special features,
and it is inevitable to consider those special features if researchers want to have more
accurate results in China. Chen (2004), Mao et al. (2008), Zhang and Xu (2013), Liu and
Yang (2010) and Hung et al. (2015) all do their research considering one or several special
features on Chinese stock market. We summarize three primary features which are also
most frequently employed by literature.
- Firstly, it is well known that China, like many markets in the Asian region, has
substantial holdings of non-traded shares which mean that these shares are not
effectively valued. Before April 2005, listed companies had two kinds of shares
outstanding which are tradable shares and non-tradable shares. Non-tradable shares
were held by government agencies or government-related enterprises and were non-
tradable in the public market. The Chinese government started the share-structure
reform in April 2005 to legally convert non-tradable shares to tradable shares.
Almost all Chinese listed companies completed the reform by the end of 2006.
Using only tradable shares or all shares to value weight stock returns is necessary to
investigate.
36
- Another important special feature is the segmentation of Chinese stock market,

more than 170 Chinese listed firms have issued multiple class shares which have the
same cash flow and voting rights but are traded in different markets. Some of them
have A-shares and B-shares, some have A-shares and H-shares and others have the
A-shares and shares in other foreign markets. Since these shares share the same cash
flow and voting rights, they usually have the same claim on the firm’s book value of
equity. Our research focus only the CNAS stock market, in order to obtain the
book-to-market equity ratio per A-share of a company with multiple class shares, it
is incorrect to divide the firm’s total book value equity from its balance sheet by the
total market value. Instead, the correct way is to calculate the book value of equity
per share divided by the A-share price.
- Thirdly, China has two main boards for the firms to go public, the SSE and the
SZSE. In addition, the SME and the GEB were set up in May 2004 and October
2009, respectively, and both are hosted by the Shenzhen Stock Exchange. Fama and
French use NYSE-listed firms to determine the breakpoints between small and big
firms in order to avoid the overwhelming influence of the large number of small
NASDAQ firms. Therefore, whether SME and GEB listed firms should be excluded
in determining the breakpoints for the size factor in China need to be examined.
Zhang and Xu (2013) conclude in their paper that the SME and GEB stocks are
included or excluded from the sample to divide firms into size groups do not have a
distinct difference.
Based on these special features of Chinese stock market, Zhang and Xu (2013) construct FF
three factors and process the regressions separately with and without these special features.
They come to the conclusions that the performance of FF3F Model is better when the non-
tradable shares are excluded from the sample and when the book-to-price ratio (B/P) are
used instead of the B/M ratio.
On account of the special features of Chinese stock market study in literature, we construct
value-weighted stocks by their tradable shares, use B/P ratio instead of B/M ratio, and
construct of size factor by the total market capitalization including SME and GEB in our
following research.
37
1.4 Empirical Analysis and Results of FF3F Model
1.4.1 Data and methodology
1.4.1.1 Data
All the firms on CNAS stock market, excluding financial firms18 and firms with negative
market-to-book values, are collected from Shanghai A-Share Index and Shenzhen A-Share
Index. In addition, a firm is eliminated if the relevant information is missing in a particular
month or period, and the obvious errors are corrected manually.
For the period of July 2004 to May 2015 (131 months), monthly index prices and stock
prices are obtained from Bloomberg, so as to their market capitalization, book value per
share, total shares outstanding and listed shares outstanding. Furthermore, risk-free rate (RF
rate) is a typically proxy for the return on a one-month Treasury bill. But in China, the one-
month Treasury bill has never been issued until February 2007. To keep it consistent with
our sample period, we replace it with ‘Three-Month Treasury Bill Rate (3M T-Bill Rate)’
and the one-month risk-free rate is then equal to the 3M T-Bill Rate divided by three.
Considering the special features of Chinese stock market, and as already demonstrated in
the section 1.3.2 that the best performance of three-factor model is achieved when the three
factors are constructed by using the market portfolio that includes only tradable shares;
using the total market value to determine size breakpoints; and when the B/P ratio is used
instead of the B/M equity ratio.
For each stock:
- The stock return is defined as the logarithm of the difference of monthly price.
- The firm’s size is defined as the natural logarithm of the market capitalization in
local currency Chinese ‘yuan’ (RMB).
- Instead of B/M equity ratio, B/P ratio is calculated as a firm’s book value per share
divided by its price.
- To calculate the value-weighted returns, the tradable market value of equity instead
of the total market value of equity is used, the tradable market value of equity
18
The financial firms are excluded because their high leverage which is normal for these firms
maybe does not have the same meaning for non-financial firms, where high leverage more likely
presents high distress risk.
38
equals to the stock price of each month times the number of tradable shares (Listed
shares outstanding).
1.4.1.2 Construction of Fama-French portfolios on Chinese market
We follow FF method as demonstrated in section 1.2.1 and construct the 6 Size-B/P

portfolios on Chinese stock market, SL, SM, SH, BL, BM, and BH, to form the size factor
SMB, at the end of June of each year t, all the stocks are sorted into two size groups, Small
and Big, the breakpoint is the median total market capitalization including SME and GEB
listed firms. For the value factor HML, we sort by the B/P ratio instead of B/M ratio at the
end of each December of year t-1 into three groups: Low, Medium and High. The
breakpoints are the 30th and 70th A-shares percentiles. Finally, the intersection of these six
groups makes the six portfolios, which remain the same from July of year t to June of year
t+1, and the portfolios are reformed in July of year t+1.
The “market” portfolio return series, which covers both the Shanghai and Shenzhen
exchanges, is not readily available. We first compute the monthly returns of Shanghai and
Shenzhen composite A-share indexes separately, then compute the value-weighted average
of the returns using the relative (aggregate) market values of the two exchanges observed at
the end of each month as weights. The weighted average market returns thus obtained are
used as our proxy for market returns.
Then the market factor is the excess market return which is computed as the difference
between the value-weighted returns of all A-shares and the RF rate. The SMB factor is then
the difference between the simple average of the monthly returns of the three small-size
portfolios (SL, SM, and SH) and the simple average of monthly returns of the three big-size
portfolios (BL, BM, and BH). Similarly, HML is equal to the simple average monthly
return of the two portfolios with high book-to-market equity (SH and BH) minus which of
the two portfolios with low book-to-market equity (SL and BL)
Most literature including FF themselves construct 25 portfolios which are the intersection
of five portfolios formed on size (Small, 2, 3, 4, and Big) and five portfolios formed on
B/M ratio (Low, 2, 3, 4, and High). The 25 Size-B/P portfolios are formed much like the
six Size-B/P portfolios discussed above, we firstly sort firms into five size groups at the end
of June of year t, the breakpoints are all the A-share market equity quintiles. Then
independently, we sort firms into five B/P ratio groups at the end of year t-1, for which the
breakpoints are all A-shares quintile. Similarly, the portfolios remain unchanged from July
of year t to June of year t+1, and the portfolios are reconstructed at June of year t+1.
39
Thus, to be included in our data set, a stock must have market equity data for December of
t-1 and June of t, and (positive) price data for t-1, in addition, firms must have monthly
returns for at least 24 out of 131 months between July 2004 to May 2014. Ultimately, the
total number of stocks which meet our selecting criteria on A-share market is 2267.
1.4.2 Empirical results on Chinese A-share stock market
1.4.2.1 Summary statistics
Table 1.2 shows the annual average available number of firms which have the data of
market capitalization, B/M ratio, and tradable market value each year without the financial
firms and negative B/M firms. The data are not so satisfactory before 2003 after the non-
financial firms and firms with negative B/M ratio are eliminated from the sample. The
available number of B/M ratio are always less than that of size, especially before 2003,
such as the available number of firm size is 895 in the year 2001 while the number of firm
B/M ratio is only 15.
Table 1.2 Annual average available number of firms that have required data (2001-2014)
This table represents the annual average number of firms which have available data of market
capitalization, book-to-market equity ratio and tradable market value. The first column is the year
from 2001 to 2014, the following columns are the available amount of data of market capitalization,
B/P ratio and tradable market value of A-share stock market, separately. The symbol ‘-’ indicate the
data that is not available.
Year Market capitalization B/P ratio Tradable market value
2001 895 15 -
2002 950 476 -
2003 1013 709 352
2004 1105 929 1128
2005 1158 1020 1119
2006 1175 1008 1159
2007 1254 1106 1255
2008 1352 1218 1360
2009 1408 1286 1478
2010 1662 1500 1786
2011 1977 1846 2058
2012 2189 2069 2204
2013 2248 2110 2172
2014 2224 2040 2113
40
Besides, the number of firms which have the available tradable market value is less than
400 before 2004. To reduce the bias of the regression results, we choose the time interval
July 2004 to May 2015 and construct three factors during 131 months.
To construct the three factors, the monthly price of the market portfolios (SSE A-Share
Index and SZSE A-Share Index) and the eligible stocks, total market capitalization, tradable
market value, and book value per share of CNAS market are obtained from Bloomberg.
The 3-month Treasury Bond Trading Rate, which downloaded from Bloomberg as well,
and the monthly risk-free rate is the 3M T-Bill Rate divided by three.
Table 1.3 reports the statistic description of FF six Size-B/P ratio portfolios. The five parts:
annual average of firm size, annual average of B/P ratio, average of annual percent of
market value in portfolio, average of annual number of firms in portfolios and average
excess returns are represented in Panel A, B, C, D and E respectively. Firstly, across the
two size groups, high B/P ratio portfolios have relatively bigger size, and then across the
three B/P ratio groups, big size portfolios have relatively higher B/P ratio than small size
portfolios. In Panel C, the big size portfolios contribute about 86.30% in total with the
nearly number of firms as the small size portfolios (733 firms for small size portfolios and
732 firms for big size portfolios).
Table 1.3 Descriptive statistic of Fama-French six value-weighted Size-B/P portfolios (period:
July 2004- May 2015)
This table reports the statistic description of six Size-B/P portfolios (SL, SM, SH, BL, BM and BH),
their annually average of firm size and B/P ratio are presented in Panel A and Panel B separately,
Panel C is annual average percentage of market value in portfolios, Panel D is the annual average of
firm numbers and Panel E is the average excess return of the six portfolios. Across the columns are
the two size groups (Small and Big) and across the rows are three B/P groups (Low, Medium and
High). The unit of market value is 100 million ‘yuan’.
Book-to-Price (B/P) ratio
L M H L M H
Panel A: Average of annual firm size Panel B: Average of annual B/P ratio
S 2171 2216 2351 0.2105 0.3680 0.5695
B 11955 14265 16373 0.2294 0.3820 0.6358
Panel C: Average of annual percent of Panel D: Average of annual number
market value in portfolio of firms in portfolio
S 4.39% 4.54% 4.78% 190 316 227
B 24.74% 30.73% 30.83% 249 271 212
Panel E: Average excess returns
S 0.0236 0.0251 0.0244
B 0.0161 0.0142 0.0123
41
The average excess returns in Panel E are consistent with FF, in which big size portfolios
tend to have lower excess returns than small size portfolios. However, the average excess
returns across the B/P ratio seem not to have a clear tendency, only in the big size groups,
the excess returns decrease with B/P ratio, which is opposite to U.S. stock market.
In Table 1.4, Panel A displays the summary statistics of FF three factors (excess market
return, SMB and HML) on CNAS stock market. Panel B and Panel C shows the correlation
coefficients among dependent variables (FF three factors) and correlation coefficients
between independent variables (excess returns of six Size-B/P portfolios) and dependent
variables. Panel B indicates that the correlation coefficients between market beta and size
factor SMB is 0.0014, and between the market factor and value factor HML is 0.1857,
which are both low. The correlation coefficients between SMB and HML (-0.3224)
indicates that size factor and value factor are negatively correlated. In Panel C, all the six
portfolios and market factor are positively highly correlated. Small size portfolios are all
Table 1.4 Summary statistics of Fama-French three factors and correlation coefficients
among variables (July 2004- May 2015)
Panel A is the summary statistics of FF three factors: the excess market return, Rm-Rf; the size
factor SMB and the value factor HML. Panel B is the correlation coefficients among FF three
factors, and Panel C is the correlation coefficients between excess returns of six Size-B/P portfolios
and FF three factors.
Panel A: Summary statistics of Fama-French three factors
RM ,t  R f SMB HML
Mean -0.0006 0.0102 -0.0015
Standard Error 0.0077 0.0035 0.0028
Median 0.0051 0.0124 -0.0046
S.D. 0.0884 0.0403 0.0316
Sample Variance 0.0078 0.0016 0.0010
Kurtosis 1.1271 2.2564 4.5290
Skewness -0.6384 -0.6685 0.4895
Panel B: Correlation coefficients among Fama-French three factors
Rm-Rf 1
SMB 0.0014 1
HML 0.1857 -0.3224 1
Panel C: Correlation coefficients between dependent variables and independent variables
SL-Rf 0.8037 0.5342 -0.0460
SM-Rf 0.8233 0.4989 -0.0024
SH-Rf 0.8274 0.4674 0.1130
BL-Rf 0.8990 0.1646 -0.0951
BM-Rf 0.9320 0.1270 0.1372
BH-Rf 0.9084 -0.0137 0.4570
42
relatively highly correlated with size factor SMB (0.5342, 0.4989 and 0.4674), while the
correlation coefficients between big size portfolios and SMB are low (0.1646, 0.1270 and -
0.0137), all the six portfolios are positively related with size factor except the BH portfolio.
Furthermore, the correlation coefficients between six portfolios and value factor HML are
relatively low also except for the BH portfolio (0.4570).
Table 1.5 Descriptive statistic of Fama-French 25 Size-B/P portfolios (July 2004- May 2015)
This table displays the descriptive statistic of FF 25 SBP portfolios for the period of July 2004 to
May 2015 (131 months, 11 years). Across the columns of each panel are the five size portfolios
which from Small to Big, and across the rows are the five B/P portfolios which from Low to High.
Panel A is the value-weighted average excess returns of FF 25 Size-B/P portfolios, Panel B, C, and
D are the annual average size, B/P ratio and firm numbers. The unit of size is 100 million ‘yuan’.
Book-to-price (B/P) ratio
Low 2 3 4 High
Panel A: Average excess returns
Small 0.0123 0.0185 0.0173 0.0175 0.0172
2 0.0091 0.0128 0.0109 0.0111 0.0106
3 0.0055 0.0077 0.0090 0.0091 0.0089
4 0.0072 0.0052 0.0071 0.0067 0.0036
Big 0.0048 0.0029 0.0017 0.0025 0.0006
Panel B: Annual average size
Small 1100 1189 1187 1207 1204
2 1880 1879 1881 1880 1893
3 2810 2808 2781 2807 2788
4 4731 4693 4693 4580 4658
Big 18937 22764 25721 25807 28974
Panel C: Annual average B/P ratio
Small -0.0066 0.2935 0.3881 0.5018 0.7094
2 0.0895 0.2937 0.3859 0.5020 0.7540
3 0.1474 0.2916 0.3879 0.5018 0.7385
4 0.1704 0.2866 0.3869 0.4992 0.7714
Big 0.1653 0.2874 0.3862 0.4995 0.7869
Panel D: Average number of firms
Small 67 51 62 69 45
2 44 59 64 62 63
3 48 61 60 62 67
4 60 65 61 56 61
Big 78 63 52 49 62
43
Similar as Table 1.3, Table 1.5 represents the descriptive statistics of FF 25 Size-B/P
portfolios on CNAS market for the period July 2004 to May 2015. Panel A shows the
average excess returns of 25 portfolios, annual average size, B/P ratio and firm numbers of
the 25 portfolios are shown in Panel B, C, and D respectively.
Evidence from Panel A is consistent with both FF findings and Table 1.3, across each B/P
quintile, the average excess returns decrease with the size of portfolio increases. Still, there
seems no regular variation across each size quintile. However, it is not so clear as six Size-
B/P portfolios that higher B/ P ratio portfolios have relatively bigger size, and big size
portfolios have relatively higher B/P ratio than small size portfolios. In Panel B, we cannot
tell the difference except the biggest size quintile, in which the size of portfolios increases
with B/P ratio. There is no obvious size difference across the B/P quintiles as showed in
Panel C. Firm number of portfolios are relatively even, the portfolio of the biggest size and
lowest B/P ratio has the most number of firms (78).
1.4.2.2 Time-series regressions
Table 1.6 shows the regression results obtained from the TSR of excess returns of six
value-weighted Size-B/P portfolios on FF three factors, excess market returns, SMB and
HML using a firm’s tradable market value as a portfolio weight in computation of value-
weighted returns and using all the market value to determine the breakpoints of size and
B/P ratio groups. In Table 1.6, The left part of the table is the coefficients obtained from the
regressions, a is the intercept, b , s and h are the regression coefficients s of FF three
factors separately, and adjusted R-square. Correspondingly, the right part of the table is t-
statistics and the standard error of estimation. Numbers in bold are the t-statistics which are
significant at 5% confidence level. Coefficients of the market factor and SMB are all
significant at the 5% confidential level, both are highly and positively related to stock
returns, and three out of six coefficients are significant at the 5% level for HML factor. The
adjusted R-squares are around 0.9 with averaged value 0.9081. The standard errors of
estimation are relatively low and close to zero, which also means the satisfactory results.
Overall, the regression results demonstrate that the FF3F explains most time-series
variations of average excess stock returns on CNAS stock market during the research
period.
According to the regression results, no one can ignore the significant role of the market
factor in explaining Chinese stock returns. FF find that size and B/M equity are related to
average expected portfolio returns (size is negatively related to averages expected returns
and B/M ratio is positively related to expected returns), similarly on CNAS stock market
44
(Table 1.6), small firms have persistently higher returns (with regression coefficients of
1.2461, 1.1939 and 1.2133 for three small size portfolios) than big size firms (with
regression coefficients 0.1669, 0.2867 and 0.1997). And across the size groups, firms with
higher B/P ratio have higher returns and lower B/P ratio firms have relatively poor even
negative returns. Besides, all the intercepts of six portfolios are not indistinguishable from
zero, that is to say, FF three factors cannot completely capture the variation of excess
returns.
Table 1.6 Time-series regression of six value-weighted Size-B/P portfolios on Chinese A-share
stock market (July 2004- May 2015, 131 months)
The time-series regression results of six value-weighted Size-B/P portfolios on FF3F Model are
displayed in this table. Across the columns are the two size groups (Small and Big) and across the
rows are the three B/P ratio groups (Low, Medium and High). The left part of the table reports the
coefficients obtained from the time-series regressions and adjusted R-square. Correspondingly, the
right part of the table is t-statistics corrected for heteroscedasticity and autocorrelation using the
Newey-West estimator, and the standard error of the estimation. Numbers in bold are the t-statistics
which are significant at 5% confidence level.
Regression: Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ei ,t
L M H L M H
a t (a )
S 0.0113 0.0135 0.0132 4.8608 5.5340 5.1820
B 0.0140 0.0118 0.0122 5.6672 4.0914 4.9767
b t (b )
S 0.8766 0.8961 0.8767 33.6953 33.6442 25.7726
B 0.8469 0.9556 0.8468 29.8677 22.6775 37.0579
s t (s )
S 1.2461 1.1939 1.2133 16.8556 14.1879 13.7004
B 0.1669 0.2867 0.1997 2.1578 2.5084 2.8689
h t (h)
S -0.0818 0.0187 0.3888 -0.8383 0.1912 3.2870
B -0.6096 0.0157 0.9197 -6.1817 0.1125 10.7208
Adj. R-square Residual standard error
S 0.9290 0.9239 0.9140 0.0256 0.0266 0.0283
B 0.8831 0.8818 0.9167 0.0271 0.0312 0.0254
Since the six portfolios can be criticized to have more bias and the results may have some
coincidences. To be more convincing, and following what FF do in their paper of 1993, we
construct 25 value-weighted Size-B/P portfolios on CNAS stock market using the same
data sample. The TSR results of 25 Size-B/P portfolios are presented in Table 1.7.
45
46
Table 1.7 Time-series regressions of 25 value-weighted Size-B/P portfolios on FF3F Model, Chinese A-share stock market (July 2004
- May 2015; 131 months)
This table presents the time-series regressions results of 25 value-weighted Size-B/P portfolios on FF3F Model, Across the columns are five
size groups and across the rows are five B/P groups. The left part of the table is the coefficients of the regressions and adjusted R-square.
Correspondingly, the right part of the table is t-statistics corrected for heteroscedasticity and autocorrelation using the Newey-West estimator,
and the standard error of the residuals. Numbers in bold are the t-statistics which are significant at 5% confidence level.
Size B/P ratio

Low 2 3 4 High Low 2 3 4 High
a t (a )
Small -0.0044 0.0024 0.0019 0.0019 0.0026 -1.5841 0.8853 0.8117 0.7198 0.9889
2 -0.0047 -0.0005 -0.0034 -0.0020 -0.0025 -1.5021 -0.1645 -1.3339 -0.7461 -0.9120
3 -0.0078 -0.0044 -0.0024 -0.0023 -0.0019 -2.5702 -1.6193 -0.9183 -0.8930 -0.7440
4 -0.0022 -0.0044 -0.0019 -0.0022 -0.0048 -0.7150 -1.5121 -0.7033 -0.7121 -1.7385
Big 0.0010 -0.0004 -0.0008 -0.0001 -0.0008 0.3856 -0.1371 -0.2218 -0.0480 -0.2678
b t (b )
Small 0.9980 0.9790 0.9632 0.9921 1.0223 20.9502 24.2296 30.0813 21.0168 25.6580
2 1.0173 1.0115 1.0145 1.0217 0.9786 23.1302 23.3347 24.8023 19.5580 19.5363
3 1.0474 1.0065 1.0226 1.0397 1.0199 14.2400 16.9243 25.6594 23.3416 21.1499
4 0.9839 1.0161 1.0410 1.0642 1.0315 19.2615 23.6557 20.3761 21.3104 19.0558
Big 0.9902 1.0848 1.1129 1.0892 1.0192 19.0456 21.8792 17.4111 24.0630 20.2916
s t (s )
Small 1.7215 1.6262 1.5829 1.6192 1.5391 18.2026 17.7723 20.6765 15.6372 15.2212
2 1.3944 1.3416 1.4422 1.3500 1.4211 12.2720 15.2149 14.7213 11.7015 14.0101
3 1.3131 1.2031 1.1731 1.2051 1.1870 10.3210 10.2878 12.3951 12.7540 14.5117
4 0.8990 0.9382 0.9173 0.9591 0.9775 8.3352 8.5334 8.4023 9.5668 9.5371
Big 0.2987 0.3186 0.3068 0.3811 0.3327 2.7596 2.9922 2.1507 3.1433 2.8028
Table 1.7 Continued
Size B/P ratio

Low 2 3 4 High Low 2 3 4 High
h t (h)
Small 0.1368 -0.0191 0.1010 0.2561 0.3283 0.9822 -0.1547 1.0269 1.7704 2.7737
2 -0.1329 -0.0968 -0.0695 0.0646 0.5187 -0.8079 -0.6982 -0.6113 0.5000 3.8859
3 -0.3549 -0.2719 -0.0443 0.2148 0.4639 -1.8389 -1.7835 -0.3506 1.5733 4.0028
4 -0.5533 -0.4044 -0.1631 0.1485 0.5812 -4.3698 -3.2740 -1.2538 1.0347 4.6418
Big -0.8710 -0.4303 -0.0341 0.3459 0.9049 -6.1962 -2.8499 -0.1969 2.1630 6.6817
Small 0.8922 0.9282 0.9333 0.9122 0.9172 0.0389 0.0302 0.0284 0.0339 0.0330
2 0.9039 0.9100 0.9288 0.9065 0.9093 0.0346 0.0329 0.0296 0.0338 0.0330
3 0.8759 0.8924 0.8984 0.9063 0.9169 0.0403 0.0353 0.0343 0.0335 0.0311
4 0.8633 0.8907 0.8843 0.8859 0.9044 0.0378 0.0342 0.0358 0.0366 0.0331
Big 0.8670 0.8880 0.8747 0.8957 0.8949 0.0347 0.0340 0.0374 0.0339 0.0339
47
Chapter1 Fama-French Five-Factor Model vs. Fama-French Three-Factor Model in China
Similar as the TSR on FF six Size-B/P portfolios, the market factor still plays an important
role in explaining the portfolio excess returns, as shown in each portfolio, all the
coefficients of the market factor are dramatically significant. Regarding the size factor
SMB and value factor HML, in accordance with (Fama and French, 1993) findings, that the
average excess returns decrease as the firm size increase, the t-statistics (t-stats hereafter) of
regression coefficients for the size factor are all significant at 5% confidence level. The
excess returns have a positive relationship with firms’ B/P ratio, firms with higher B/P ratio
tend to have higher excess returns. The top 20% B/P ratio group has the most numbers of
significant coefficients with t-stats 2.7737, 3.8859, 4.0028, 4.6418 and 6.6817 respectively,
while the medium 20% group has none of the coefficients significant; and the bigger size
portfolios seem to have more significant regression coefficients on HML. The adjusted R-
squares are around 0.9 with averaged value 0.8995, the intercepts are all indistinguishable
from zero except the portfolio in the third size quintile and the lowest B/P quintile in this
regression, which means FF3F model captures the time-series variations of excess returns
well on CNAS stock market.
The size and value effects on CNAS stock market are clearly shown in Figure 1.5. Figure
1.5 reports the TSR loadings on FF three factors (b, s, h) of the 25 Size-B/P portfolios
(loadings are also reported in Table 1.7), separately. On the x-axis are the 25 portfolios
from S1H1
Figure 1.5 Loadings of 25 Size-B/P portfolios on FF3F
This figure represents the time-series regressions loadings on FF three factors of 25 value-weighted
Size-B/P portfolios – b, s, h, separately. On the x-axis are the 25 portfolios from S1H1 (smallest
size and lowest B/P ratio) to S5H5(biggest size and highest B/P ratio), noting that in the middle
panel, five portfolios of the same B/P quintile are grouped together in the order of increasing size,
whereas in other two panels, five portfolios of the same size quintile are grouped together in the
order of increasing B/P ratio.
48
Figure 1.5 Continued
(smallest size and lowest B/P ratio) to S5H5(biggest size and highest B/P ratio), noting that
in the middle panel, five portfolios of the same B/P quintile are grouped together in the
order of increasing size, whereas in other two panels, five portfolios of the same size
quintile are grouped together in the order of increasing B/P ratio. The three panels in Figure
1.5 are the loadings on excess market return (b), on SMB (s) and on HML (h). The loadings
on SMB (s) show a clear monotonically decreasing relationship with size and the loadings
on HML show a clear monotonically increasing relation with B/P ratio.
49
1.4.2.3 Cross-sectional results on Chinese stock market
The main contribution of FF3F Model is its success in explaining the cross-section stock
returns in U.S market and many other stock markets in the world. We also examine whether
FF3F Model can explain the cross-section variations of stock returns on CNAS stock
market, both in the frame of 6 Size-B/P portfolios and 25 Size-B/P portfolios. FF apply
(Fama and MacBeth, 1973) two-stage approach to perform the regressions, the first stage is
the time-series which already demonstrated in previous research. The second stage is the
cross-section regressions as shown in equation (1.16), where they use the estimated betas
(in this research b̂ , ŝ and ĥ ) which are obtained from the first stage of TSR, as the
independent variables in the CSRs. And regress the same portfolios’ returns as in the first
stage on these estimated betas for a fixed time period to determine the risk premium for
each factor.
Ri ,t  R f  i   M bî   SMB ˆsi   HML hî  i ,t (1.16)
Where:
Ri ,t  R f is the excess returns of same portfolios as in the time-series (value-weighted FF

six or 25 Size-B/P portfolios);
b̂i , ŝi and ĥi are the estimated coefficients from TSRs for the market factor, size factor
SMB and value factor HML, separately;
 M ,  SMB and  HML are the coefficients of CSRs.
Furthermore, it is important to recognize the classical Errors-in-variables (EIV) problem

before we perform the CSRs, which occurs from the nature of Fama and MacBeth (1973)
two-stage approach. When we applying standard OLS formulas to the second step of CSRs,
the independent variables ̂ s are assumed to be given; however, ̂ s are estimated from the
first step of TSRs, which are not fixed. Though Fama-Macbeth (FM) reduce the
measurement error by running CSRs for portfolios instead of individual stocks, Shanken
(1992) argues that the sampling errors for  s associated with the estimated betas still
cannot be ignored. Shanken proposes that FM approach for computing standard errors
keeps overstated the precision of the gamma estimates. Following Shanken (1992), we
apply a correction procedure to solve the EIV problem. It assumes that the error terms from
the TSRs are independently and identically distributed over time and independent of the
factors. (Details of Shanken correction is presented in Appendix A)
50
According to our database, the cross-section regressions are performed at each month (131
months, from July 2004 to May 2015), for both FF six value-weighted Size-B/P portfolios
and 25 value-weighted SBP portfolios, then the simple average of constants and
coefficients, which are obtained from the 131 times CSRs on estimated betas ( b , s and h )
are calculated.
Table 1.8 shows the CSRs results on the estimated coefficients which obtained from the
TSRs of FF 6 SBP value-weighted portfolios (Table 1.6) and 25 Size-B/P value-weighted
portfolios (Table 1.7) separately. Panel A presents the CSR results of six value-weighted
Size-B/P portfolios, and Panel B shows the CSR results of 25 value-weighted Size-B/P
portfolios.  M ,  SMB and  HML are the coefficients of the CSRs and their corresponding
Fama-MacBeth (FM) and Shanken (SH) corrected t-stats are presented in the parentheses
below the gammas.
The CSR results of FM show that neither of  M are significant at 5% confidence level (-
0.0116 for the six portfolios with t-stats -0.6678, and -0.0251 for the 25 portfolios with t-
stats -1.5912), which is consistent with most of researches on Chinese stock market that
market beta is not able to explain the cross-sectional stock returns. Both  SMB of the two
CSRs are
Table 1.8 Cross-sectional regressions of six value-weighted Size-B/P portfolios and 25 value-
weighted Size-B/P portfolios, Chinese A-share stock market (July 2004- May 2015)
This table presents the results of cross-section regressions on FF3F Model of FF six value-weighted
Size-B/P portfolios (Panel A) and 25 value-weighted Size-B/P portfolios (Panel B). In each panel,
the first row is the cross-sectional regressions’ coefficients (coef.); the second row is the
corresponding Fama-MacBeth t-statistics (FM t-stats) at 5% confidence level and the third row is
the Shanken corrected t-statistics (SH t-stats), which are in the parentheses. The numbers in bold are
the t-stats which are significant at 5% level. The adjusted R-squares are percentage values.
Regression: Ri ,t  R f  i   M bî   SMB ˆsi   HML hî  i ,t
 M  SMB  HML Adj. R 2 (%)
Panel A: Cross-sectional regressions of six Size-B/P portfolios
gamma (coef.) 0.0224 -0.0116 0.0102 -0.0024 68.40
FM t-stats (1.6619) (-0.6678) (2.8861) (-0.8454)
SH t-stats (-1.4218) (2.8861) (-0.8404)
Panel B: Cross-sectional regressions of 25 Size-B/P portfolios
gamma (coef.) 0.0252 -0.0251 0.0085 -0.0004 49.01
FM t-stats (1.7655) (-1.5912) (2.2805) (-0.1200)
SH t-stats (-3.0383) (2.2882) (-0.1188)
51
significant (0.0102 for the six portfolios with t-stat 2.8861 and 0.0085 for 25 portfolios with
t-stats 2.2805), which proves the significant positive size premium of cross-sectional stock
returns. After corrected by the Shanken correction, the SH t-stats is significant (-3.0383) for
the market beta; while the important size premium is robust to the EIV adjustment.
However, neither the FM or the SH t-stats indicate that there exists significant value
premium. FF3F Model can explain more cross-sectional variation of six portfolios (with
average adjusted R-squares 68.40%) than 25 portfolios (with average adjusted R-squares
49.01%) on Chinese stock market.
Consistent with most of previous studies, such as Wang and Xu (2004), Eun and Huang
(2007), Wang and Iorio (2007) and Chen et al. (2015 working paper), the results of cross-
section regressions in Table1.8 continue to prove that there is significant positive size
premium across the stock returns on CNAS stock market during the periods July 2004 to
May 2015. However, contrary to some researches, such as Wong et al. (2006), Eun and
Huang (2007), Wang and Iorio (2007), Wu (2011) and Gan et al. (2013), we find the lack
of value premium on CNAS stock market during our research period. Market beta is able to
explain the cross-sectional variation in the average stock returns for the 25 value-weighted
SBP portfolios with negative SH adjusted t-stats.
1.4.3 Comparing with U.S. market (FF3F Model)
Since the Chinese stock market has several special features that we take into account in our
empirical research, it is important and interesting to compare the empirical results of
Chinese stock market with those of U.S. market. We perform the same regressions with the
same time interval from July 2004 to May 2015 using U.S. data. The FF three factors and
the FF six (and 25) value-weighted Size-B/M portfolios are downloaded directly from the
website of Kenneth R. French. All the empirical results of U.S. stock market are presented
in Appendix B.
Table B.1 and Table B.2 reports the TSR results of FF3F Model on U.S. market during the
period July 2004 to May 2015 (to be consistent with the time interval that we implement
empirical analysis on Chinese stock market), respectively by regressing FF six value-
weighted Size-B/M portfolios and 25 value-weighted Size-B/M portfolios on FF three
factors, excess market return, SMB and HML.
We compare Table B.1 with Table 1.6, while Table B.2 with Table 1.7. It is observed that
the time-series results on both countries are quite close. In Table B.1, all the loadings on
52
market beta and five out of six loadings on SMB are statistically significant at 5%
confidence level; while in Table B.2, still, all the loadings on market beta and 24 out of 25
loadings on SMB are significant. Take Table B.2 for example, more significant loadings
(18 out of 25 on U.S. market, while 12 out of 25 on Chinese market) on HML of U.S.
market indicates that the value factor has more explanatory power for the time-series
variation of U.S. stocks than that of CNAS stocks.
There exist size and value effect on both CNAS stock market and U.S. stock market.
Within each B/M group, the loading on SMB decreases as portfolio size increases, on U.S.
market, there are even negative loadings of the portfolios that have big size; and within
each size group, we find a positive relationship between the loading on HML and B/M ratio.
Comparing the adjusted R-squares of Table B.2 and Table 1.7, which are higher on U.S.
market (averaged adjusted R-squares is 0.9396) than on Chinese market (averaged adjusted
R-squares is 0.8995). The results reveal that the time-series variation of stock returns
captured by FF3F Model in U.S. is more than that captured by FF3F Model in China, that is
to say, FF3F Model performs better on U.S. stock market than on CNAS stock market
during the sample period.
We also perform the CSRs on FF3F Model using U.S. data, and the results are reported in
Table B.3 (Appendix B). It is interesting that none of the regression coefficients on
loadings of market beta, SMB and HML is significant; the results are robust to the EIV
problem except the market beta for the six portfolios. For the 25 value-weighted Size-B/M
portfolios, FF3F Model does not have any explanatory power of the cross-sectional
variation of stock returns on U.S. market over the period July 2004 to May 2015. FF3F
Model performs better in capturing cross-sectional variation of average excess stock returns
on CNAS stock market (with averaged adjusted R-square 68.40% of the six portfolios and
49.01% of the 25 portfolios) than on U.S. stock market (with adjusted R-square 58.14%
and 40.76% for the six and 25 Size-B/P portfolios separately) over the research period.
1.5 Construction of profitability and investment factors and empirical results of

FF5F Model
1.5.1 Construction problems
Table 1.9 shows the annual available number of firms which have available OP and Inv
data from 2004 to 2014. The number of firms that has available data of OP are always less
53
available than that of Inv, and before 2009, there are few firms (less than 30) that have
available OP.
Table 1.9 Annual firm numbers that have available data of OP and Inv (2004-2014)
This table presents the annual number of firms that have available OP and Inv data from 2004 to
2014.
Year 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
OP 12 17 24 26 27 131 294 392 777 1043 2417
Inv 1154 1237 1346 1402 1624 1981 2241 2355 2361 2525 2525
The annual available number of firms that has available OP and Inv are visually displayed
in Figure 1.6, the blue bar is operation profitability and the red bar is investment
opportunity, the x-axis indicates the year and the y-axis indicates the firm numbers. It is
obvious that the number of firms that has OP are much less than that has Inv until the year
2014. And there are few available data of OP before 2010. To be more accurate and reduce
the bias generated because of the very few firm numbers (when we sort firms into
portfolios, there may be no firms in portfolios with firm numbers less than 30 in a year), we
exclude the data before 2010.
Figure 1.6 Annual number of firms that has available data of OP and Inv
FF (2015a) perform the TSR using 25 Size-B/M Portfolios, 25 Size-OP portfolios and 25
Size-Inv portfolios. Following the same method, firstly we construct these portfolios on
CNAS stock market. The construction of Size-OP portfolios and Size-Inv portfolios are
similar as the method to construct 25 Size-B/P portfolios on Chinese stock market (refer to
the section 1.2.1.1), just the B/P ratio is replaced by OP or Inv.
Table 1.10 shows the annual number of stocks in 25 Size-OP portfolios and 25 Size-Inv
portfolios, in which, S indicates the size group, P is the profitability groups and I is the
investment groups. For instance, S1P1 portfolio indicates the intersection of firms in the
54
1.5 Construction of profitability and investment factors and empirical results of FF5F Model
bottom 20% size quintile and firms in the bottom 20% OP quintile. It is obvious that there
are more stocks of Size-Inv portfolios than that of Size-OP portfolios. Furthermore, there
are no firms in several Size-OP portfolios (portfolio S1P2 of the year 2010, portfolio S2P5
of the year 2010 and 2012), and all the portfolios except one (portfolio S2P1 of the year
2011) of the year 2010 and 2011 have no more than 5 firms. In this case, we use the frame
of six portfolios to test FF5F model on CNAS stock market instead of the 25 portfolios.
Table 1.10 Annual number of stocks in 25 Size-OP portfolios and 25 Size-Inv portfolios (2010-
2014)
This table presents the annual firm numbers in each 25 Size-OP portfolios (left-hand part) and 25
Size-Inv portfolios (right-hand part) from 2010 to 2014, in which, S indicates the size group, P is
the profitability groups and I is the Investment groups. For instance, S1P1 portfolio indicates the
intersection of firms in the bottom 20% size quintile and firms in the bottom 20% OP quintile.
Year
2010 2011 2012 2013 2014 2010 2011 2012 2013 2014
S1P1 1 5 27 38 167 S1I1 155 131 130 152 149
S1P2 0 4 20 39 131 S1I2 82 67 89 111 112
S1P3 1 1 7 15 86 S1I3 44 56 72 70 80
S1P4 1 1 2 6 44 S1I4 32 52 49 52 56
S1P5 3 1 2 4 21 S1I5 19 90 100 64 64
S2P1 5 8 25 41 129 S2I1 87 88 116 105 121
S2P2 2 2 17 34 113 S2I2 68 97 92 108 113
S2P3 1 2 16 23 104 S2I3 60 56 69 93 97
S2P4 2 2 3 6 73 S2I4 50 48 73 63 68
S2P5 0 4 0 5 28 S2I5 66 108 88 83 62
S3P1 5 5 30 42 88 S3I1 66 94 82 109 92
S3P2 3 8 31 42 107 S3I2 81 89 101 99 101
S3P3 3 2 7 27 104 S3I3 71 69 98 86 100
S3P4 0 0 2 5 101 S3I4 54 60 73 84 94
S3P5 1 3 3 5 45 S3I5 60 82 86 73 75
S4P1 21 25 33 45 54 S4I1 55 69 77 72 69
S4P2 18 21 37 46 83 S4I2 60 90 99 83 95
S4P3 9 17 31 44 110 S4I3 81 84 84 106 95
S4P4 9 12 9 27 126 S4I4 65 78 85 100 96
S4P5 2 4 5 10 69 S4I5 69 75 95 91 108
S5P1 27 35 40 41 42 S5I1 35 41 55 51 59
S5P2 34 43 48 45 49 S5I2 64 90 71 83 66
S5P3 42 53 75 78 72 S5I3 77 112 113 98 91
S5P4 39 62 72 78 109 S5I4 84 100 128 125 122
S5P5 31 53 46 61 166 S5I5 71 53 72 94 124
55
1.5.2 Empirical results of Fama-French Five-Factor Model on Chinese A-share stock

market
1.5.2.1 Summary information of factors and portfolios
Then we sort portfolios into six value-weighted Size-OP portfolios and six value-weighted
Size-Inv portfolios, the annual number of firms in the two sets of portfolios are displayed in
Table 1.11. The small size groups of Size-OP portfolios relatively have fewer stocks than
that of big size groups and the SR portfolio has no stocks in the year 2009 and only one
stock in SN portfolio. Therefore, because of the lack of data on firm numbers of CNAS
stock market, the interval of our research to processing FF5F model is from July 2010 to
May 2015 (59 months).
Table 1.11 Annual number of stocks in six Size-OP portfolios and six Size-Inv portfolios
This table presents the annual firm numbers of six Size-OP portfolios (Panel A) and six Size-Inv
portfolios (Panel B) from 2009 to 2014. In the first column of Panel A presents the six Size-OP
portfolios (SW, SN, SR, BW, BN and BR), and in the first column of Panel B shows the six Size-
Inv portfolios (SC, SN, SA, BC, BN and BA).
Year 2009 2010 2011 2012 2013 2014
Panel A Size-OP portfolios
SW 5 11 22 92 154 488
SN 1 7 7 52 107 483
SR 0 5 9 7 13 146
BW 34 76 95 140 157 233
BN 51 104 147 210 255 468
BR 28 57 93 87 121 403
Panel B Size-Inv portfolios
SC 328 374 392 404 457 465
SN 262 288 310 380 412 444
SA 105 167 289 314 258 244
BC 187 203 244 277 274 268
BN 301 373 468 492 496 479
BA 206 251 276 330 358 409
As shown in Table 1.12, Panel A is the summary statistics of FF five factors on Chinese
stock market, the mean, standard deviation, standard error, sample variance and so on.
Panel B is the correlation coefficients among the FF five factors, the profitability and
investment factors are both positively related to market factor with low correlation
coefficients (0.0418 and 0.1190) and negative related to size factor (-0.2227 and -0.2199).
56
RMW is negatively related to value factor HML (-0.0217), while CMA is positively and
relative highly related to HML with correlation coefficients of 0.4621. And the correlation
coefficients between RMW and CMA is -0.3121.
Table 1.12 Summary statistics of Fama-French five factors (July 2010-May 2015)
Panel A report the summary statistics of FF five factors, it summarizes the mean, standard error (Sd
error), standard deviation (S.D), variance, kurtosis and skewness. Panel B reports the correlation
coefficients among the five factors.
Panel A: Summary statistics of FF five Factors
RM ,t  R f SMB HML RMW CMA
Mean -0.0014 0.0106 -0.0059 -0.0061 0.0008
Sd error 0.0084 0.0038 0.0046 0.0036 0.0025
Median -0.0024 0.0117 -0.0075 -0.0128 0.0001
S.D 0.0646 0.0294 0.0355 0.0273 0.0196
Variance 0.0042 0.0009 0.0013 0.0007 0.0004
Kurtosis 0.2068 6.4386 5.9071 -0.4204 -0.2635
Skewness 0.1439 -1.2015 0.5658 0.3288 0.2217
Panel B: Correlation coefficients among FF five factors
RM-RF 1
SMB 0.1165 1
HML -0.0013 -0.6970 1
RMW 0.0418 -0.2227 -0.0217 1
CMA 0.1190 -0.2199 0.4621 -0.3121 1
Table 1.13 presents the average excess return of six Size-B/P portfolios (Panel A), Size-OP
portfolios (Panel B) and Size-Inv portfolios (Panel C). Across the columns are the two size
groups and across the rows are the three B/M groups, three OP groups, and three Inv groups,
respectively. It is apparent that there is the size effect, the big size portfolios always have
the lower returns than the small size portfolios in each panel. Across the OP groups in
Panel B, it is strange that the robust portfolios have lower returns than weak portfolios,
perhaps the lack of data for OP causes the bias. Across the Inv groups in Panel C, it seems
the neutral investment portfolios have the highest excess returns (0.0158 for small size and
neutral investment portfolio, 0.0050 for big size and neutral investment portfolio) than the
conservative and aggressive investment portfolios.
57
Table 1.13 Average monthly excess returns for portfolios formed on Size-B/M, Size-OP and
Size-Inv (July 2010-May 2015, 59 months)
The average excess returns of six Size-B/M portfolios, Size-OP portfolios, and Size-Inv portfolios
are presented in panel A, B and C respectively. Across the columns are the two size groups (Small
and Big) and across the rows are the three B/M groups (Low, Medium and High), three OP groups
(Weak, Neutral and Robust) and three Inv groups (Conservative, Neutral and Aggressive),
respectively.
Panel A: Excess returns of size-B/M portfolios
L M H
S 0.0236 0.0231 0.0207
B 0.0151 0.0092 0.0061
Panel B: Excess returns of Size-OP portfolios
W N R
S 0.0172 0.0170 0.0081
B 0.0046 0.0082 0.0016
Panel C: Excess returns of Size-Inv portfolios
C N A
S 0.0136 0.0158 0.0121
B 0.0033 0.0050 0.0031
1.5.2.2 Time-series regressions of FF5F model on Chinese A-share stock market
To understand how FF five factors explain the excess return of these portfolios, the TSRs
are performed on six Size-B/P portfolios, Size-OP portfolios and Size-Inv portfolios on FF
five factors for the period of July 2010 to May 2015 (59 months). The results are
demonstrated in Table 1.14, Panel A, Panel B and Panel C are the TSRs results for the
value-weighted six Size-B/P portfolios, six Size-OP portfolios and six Size-Inv portfolios,
separately. The loadings on market beta (b) are similar for the three sets of portfolios, they
are all highly significant at 5% confident level.
We next look at each panel, in Panel A, five out of six (except the portfolio of big size and
high B/P ratio) loadings on size factor SMB are significant at 5% confidence level, and the
signs of slopes indicate that portfolios of small size have returns that are positively related
to SMB, while returns of big size portfolios are negatively related to SMB. All the loadings
on HML are highly significant, there exists consistently size and value effect in the
regressions of six value-weighted Size-B/P portfolios on FF5F Model. However, none of
the loadings on
58
Table 1.14 Time-series regressions of six value-weighted Size-B/P portfolios, Size-OP

portfolios and Size-Inv portfolios on FF5F Model, Chinese A-share stock market (July 2010 to
May 2015, 59 months)
This table presents the time-series regressions results of FF5F model. In each panel, the regression
intercept a , the regression coefficients b , s , h , r and c of market factor, size factor, value factor,
profitability factor and investment factor, adjusted R square are respectively presented in the left
part of the table, the corresponding t-statistics corrected for heteroscedasticity and autocorrelation
using the Newey-West estimator and residual standard error are presented in the right part. Panel A
is the regressions on six Size-B/P portfolios, across the columns are the two size groups (Small and
Big) and across the rows are the three B/P groups (Low, Medium and High). Panel B is the
regression results of six Size-OP portfolios, same as Panel A, across the columns are the two size
groups and across the rows are the three OP groups (Weak, Neutral and Robust). Panel C is the
regression results of six Size-Inv portfolios, across the columns are the two size groups and across
the rows are the three Investment groups (Conservative, Neutral and Aggressive). Numbers in bold
are the t-stats which are significant at 5% confidence level.
Regression: Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ri RMW  ci CMA  ei ,t
Panel A Time-series regressions of six Size-B/P portfolios
L M H L M H
a t (a )
S 0.0102 0.0105 0.0108 7.4308 4.4938 5.6696
B 0.0124 0.0091 0.0118 6.8522 3.8634 6.4458
b t (b )
S 0.9637 0.9964 0.9703 41.1513 36.2183 35.8284
B 0.8361 1.0214 0.8295 27.6969 28.5669 20.8687
s t (s )
S 1.0039 0.9383 0.8557 15.9153 16.3385 11.3131
B -0.1946 -0.2434 -0.0465 -2.8370 -2.4971 -0.5165
h t (h)
S -0.5849 -0.5197 -0.2689 -6.2171 -6.9004 -3.9751
B -0.9928 -0.6007 0.6912 -12.4860 -7.2244 5.1532
r t (r)
S -0.0695 -0.1448 -0.0617 -1.1264 -1.9122 -0.7331
B 0.0188 -0.0456 0.0110 0.2597 -0.6538 0.1965
c t (c)
S 0.2515 0.1051 0.3064 2.6156 1.0264 2.3582
B 0.1114 0.2802 0.0565 1.2338 3.4851 0.5584
S 0.9782 0.9714 0.9606 0.0120 0.0137 0.0148
B 0.9625 0.9609 0.9513 0.0122 0.0136 0.0134
59
Table1.14 Continued
Panel B: Time-series regressions of six Size-OP portfolios

Operating Profitability
W N R W N R
a t (a )
S 0.0012 0.0020 -0.0018 0.5498 0.4592 -1.0778
B -0.0009 0.0028 0.0021 -0.5008 1.4503 0.6190
b t (b )
S 1.0075 1.0408 1.0492 35.8879 20.1490 31.7018
B 1.1300 1.0253 1.0883 26.7879 34.3991 25.3012
s t (s )
S 1.1712 0.9800 1.5637 13.2828 5.5382 18.1679
B 0.2480 0.2628 -0.1445 2.3517 3.3480 -1.1030
h t (h)
S -0.4482 -0.7244 -0.2020 -4.5108 -3.9157 -1.9726
B -0.4560 -0.5496 -0.7022 -4.4978 -5.9760 -6.6825
r t (r )
S -0.3429 -0.2601 1.1319 -4.6763 -2.5519 15.7233
B -0.2265 -0.1198 0.2987 -3.5011 -1.3591 3.4009
c t (c )
S 0.2644 0.1610 0.5398 2.4244 0.7310 4.2483
B 0.4613 0.0414 0.1860 5.8956 0.3561 1.3955
S 0.9720 0.9301 0.9653 0.0143 0.0238 0.0172
B 0.9643 0.9640 0.9486 0.0150 0.0139 0.0172
Panel C: Time-series regressions of six Size-Inv portfolios
Investment
C N A C N A
a t (a )
S -0.0016 0.0018 -0.0017 -0.9026 0.8186 -1.0068
B -0.0030 0.0003 -0.0029 -1.5819 0.1566 -1.5573
b t (b )
S 1.0708 1.0548 1.0274 31.3516 33.9728 35.4151
B 1.0683 1.0704 1.1116 27.9726 32.5243 27.7982
s t (s )
S 1.2837 1.1137 1.1998 18.0777 14.7519 14.5888
B 0.4139 0.3165 0.4978 4.0713 4.7070 5.5174
60
Table1.14 Continued
Panel C: Time-series regressions of six Size-Inv portfolios

Investment
C N A C N A
h t (h)
S -0.2393 -0.5369 -0.5135 -2.2282 -6.2005 -4.8437
B -0.6269 -0.4482 -0.3527 -5.9485 -5.6888 -3.2055
r t (r)
S -0.1329 -0.0789 -0.0871 -1.3784 -0.8737 -0.9804
B 0.0481 -0.0330 0.0023 0.5404 -0.5768 0.0249
c t (c)
S -0.7507 -0.0210 0.5330 -4.8137 -0.2007 3.2129
B -0.2540 0.0445 0.4623 -1.9740 0.4260 3.7475
S 0.9722 0.9713 0.9739 0.0148 0.0149 0.0141
B 0.9607 0.9680 0.9595 0.0157 0.0135 0.0160
the profitability factor RMW is significant, while three out of six loadings on the
investment factor CMA are significant at 5% confidence level.
Comparing the TSRs in Panel A with those (Appendix C, Panel A of Table C.1) of the six
value-weighted SBP portfolios on FF3F model over the same time interval (July 2010 to
May 2015), the results are quite similar for FF three factor (market beta, SMB, and HML).
The adjusted R-squares are much close between both regressions on FF3F Model and FF5F
Model, it is suggested that FF profitability and investment factors seem not add explanatory
power in capturing time-series variation of excess stock returns on CNAS stock market
during the sample period.
In Panel B, the regression results for market beta, SMB, and HML are fairly close to those
of Panel A, the big difference is in profitability factor RMW, all loadings on RMW except
the portfolio BN are significant; and in each size group, portfolios with robust profitability
tend to have higher excess returns than portfolios with weak profitability. Three out of six
coefficients of investment factor CMA are significant, two are the portfolios with weak
profitability (0.2644 for portfolio SW with t-stats 2.4244 and 0.4613 for portfolio BW with
t-stats 5.8956) and one is the portfolio SR (coefficients 0.5398 with t-stats 4.2483).
Comparing to the regressions of the same six value-weighted Size-OP portfolios on the
original FF3F Model on Chinese stock market (Appendix C, Panel B of Table C.1), in the
presence of RMW and CMA, FF original three factors do not lose their explanatory power.
61
Comparing the adjusted R-square of both regressions, FF5F Model explains slightly more
variations of time-series average returns (with averaged adjusted R-square 0.9574) than
FF3F Model (with averaged adjusted R-square 0.9256) of the six value-weighted Size-OP
portfolios.
In Panel C, the regression results of the market factor, SMB, and HML are all satisfactory
significant. The loadings on RMW are similar to Panel A, none of which is significant at 5%
confidence level. As for the CMA factor, three out of six loadings are significant,
furthermore, the three significant loadings are of the portfolios with conservative and
aggressive investment. And the investment effect is close to the results of 25 Size-Inv
portfolios of Fama and French (2015a), the aggressive investment portfolios tend to have
smaller even negative regression loadings, while the conservative investment portfolios
have relatively bigger regression loadings. In other words, there exist investment effect and
the firm’s investment is negatively related to average excess stock returns. Comparing to
the regression results of the same portfolios but on FF3F Model (Appendix C, Panel C of
Table C.1), it is suggested that RMW and CMA factors seem not add explanatory power in
capturing time-series variation of the six value-weighted Size-Inv portfolios’ returns (the
averaged adjusted R-square are 0.9658 for FF3F Model and 0.9676 for FF5F Model),
neither.
To summarize, market beta always plays an important role in explaining time-series

variation of excess portfolio returns. For all the three sets of portfolios, there exists size
effect that the excess returns are negatively related to firm size. While there exists value
effect in SBP portfolios, profitability effect in Size-OP portfolios and investment effect in
Size-Inv portfolios. The loadings on RMW are only significant in the set of portfolios
sorted by size and OP, but not in two other sets of portfolios. As to the CMA factor, the
significant loadings are concentrated in the extreme OP or Inv groups (such as the weak OP
group, robust OP group, the aggressive and conservative Inv groups). However, for the
Size-B/P portfolios, the CMA significant coefficients are relatively dispersive. In short,
whether FF5F Model performs better than FF3F Model on CNAS stock market over the
sample period is not quite clear. The explanatory power of FF5F Model seems differs
among different sets of portfolios. In comparison with FF3F Model, the presence of
profitability and investment factors seem not capture more variations of expected stock
returns than the three-factor model except for the six value-weighted portfolios formed on
size and OP, though the improvement is limited.
62
1.5.3 Comparing with U.S. stock market (FF5F Model)
Similarly, we compare the performance of FF5F Model on both CNAS stock market and on
U.S. stock market. We implement the same regressions in the previous section as reported
in Table 1.14 but using data of U.S. market. The six value-weighted Size-B/M portfolios,
six Size-OP portfolios, and six Size-Inv portfolios are downloaded directly from Kenneth R.
French’s website, the TSR results of the three sets of portfolios are presented in Appendix
B (Table B.4). The loadings on the excess market return are always strongly positive for all
three sets of portfolios of both countries. The loadings on SMB are strongly positive for
small stocks and slightly positive or negative for big stocks, there exists size effect on both
stock markets.
We next compare between each panel of Table 1.14 (Chinese market) and Table B.4 (U.S.
market). Comparing ‘Panel A’ of both tables, there exists value effect on both stock
markets. As to the profitability factor RMW, four out of six loadings on RMW are
statistically significant and especially all three loadings on small portfolios are negative
significant in U.S.; while none of the loadings on RMW is significant at 5% confidence
level in China.
Comparing Panel B of both tables, the regression results of six Size-OP portfolios are
approximately close. All the loadings on profitability factor RMW are strongly significant,
among which the loadings are strongly negative for the weak OP portfolios (low
profitability) and strongly positive for the robust OP portfolios (high profitability) on U.S.
stock market; while five out of six loadings on RMW are significant on CNAS stock
market with the same pattern as U.S. market. It is noticed that the loadings on CMA factor
are significant only for the three big size portfolios in U.S. We find no apparent value effect
when regressing the six Size-OP portfolios on FF5F Model on both stock markets.
The regression results for the six Size-Inv portfolios are quite different comparing Panel C
of both markets. First, most loadings on HML lose their significance (only one out of six is
significant) in U.S.; while all the portfolios have strong negative exposure to HML on
Chinese stock market but no value effect. Then the small size portfolios always have
significant exposure to RMW in U.S.; while none of the loadings on RMW is significant on
CNAS stock market for the Size-Inv portfolios. Last, CMA factor seems explains more
time-series variation of excess stock returns in U.S. than in China, since all the loadings on
CMA are significant while only four out of six loadings are significant on Chinese stock
market. The slopes of conservative (low investment) portfolios are positive and the slopes
of aggressive (high investment) portfolios are negative on both markets, which is consistent
with FF’s expected pattern.
63
Furthermore, the adjusted R-squares of six Size-OP portfolios (with averaged adjusted R-
squares 0.9545 on Chinese market, and 0.9861 on U.S. market) and six Size-Inv portfolios
(with averaged adjusted R-squares 0.9676 on Chinese market, and 0.9855 on U.S. market)
are slightly bigger in U.S than that in China, which indicates that FF5F Model explains the
two sets of portfolios slightly better on U.S. stock market than on CNAS stock market. In
addition, the profitability factor and investment factor are able to capture partially time-
series variation of all three sets of portfolios’ returns on U.S. stock market, while on
Chinese stock market, the profitability factor seems to be an explanatory factor only for the
six value-weighted Size-OP portfolios.
1.6 Conclusions
Fama and French draw a conclusion that risk (market beta) was not able to identify all the
stock return variations during 1963-1990 on US stock market, two other factors, size and
book-to-market equity, combined to capture the cross-sectional variation in average stock
returns unite with market β. To investigate whether size and value factors also explain
excess returns on Chinese stock market, this chapter first performs the empirical tests of
FF3F Model on CNAS stock market.
As the results shown in the empirical tests on CNAS stock market (July 2004-May 2015),
FF3F model can explain a majority of time-series variation of the stock returns, when using
tradable market value to weight the portfolios, total market capitalization to decide the size
breakpoint and book-to-price ratio instead of B/M equity ratio. We conduct the CSRs and
the results are consistent with most of the previous studies on Chinese stock market, there
exists positive size premium across the stock returns on CNAS stock market, however, we
find no value premium during the periods July 2004 to May 2015. While after adjusted by
Shanken correction, the market beta is also a determinant factor in explaining the cross-
sectional excess stock returns (25 value-weighted Size-B/P portfolios).
Despite the quite significant results of regressions on CNAS stock market, FF three factors
(size factor and book-to-market equity combined with the market factor) still cannot
explain all the variation of stock returns.
According to Fama and French, firm size and B/M equity ratio are related to the systematic
pattern of profitability and growth. They are potentially major sources of risk in return.
These two mentioned variables were known in most studies as two specific market
indicators that raise questions about the model. These findings diminished the credence of
64
1.6 Conclusions
this model, and a new wave was formed in the development field of financial theories with
the aim of explaining the causes of these special consequences.
Based on the valuation theory and recent empirical findings on the strong profitability and
investment effects in asset returns, FF (2015a) propose a five-factor model contains the
market factor and factors related to size, book-to-market equity ratio, profitability and
investment. We apply FF5F Model on CNAS stock market during the period July 2010 to
May 2015 and construct three sets of portfolios similarly as FF, six value-weighted Size-
B/P portfolios, six value-weighted Size-OP portfolios and six value-weighted Size-Inv
portfolios. For all the three sets of portfolios, market factor, size factor and value factor
have strong explanatory power for the expected excess returns in the presence of
profitability and investment factors. There always exists size effect that the excess returns
are negatively related to firm size, while there exists value effect in Size-B/P portfolios,
profitability effect in Size-OP portfolios and investment effect in Size-Inv portfolios. The
CMA factor do have explanatory power for certain portfolios in all three sets of portfolios.
However, the RMW factor only has explanatory power in six Size-OP portfolios.
In comparison with FF3F Model, profitability and Investment factors seem not having
much additional explanatory power except for the six value-weighted portfolios formed on
size and OP. The explanatory power of FF5F Model seems differs among different sets of
portfolios comparing with the original three-factor model on CNAS stock market during the
research period July 2010 to May 2015. Since the research period is relatively short in this
study, we suggest to apply the examination with a longer time interval for the FF5F Model
on Chinese stock market in the future.
We also implement the regressions that performed on CNAS stock market over the same
period using U.S. data. The empirical results reveal that both FF3F Model and FF5F Model
explain slightly better time-series variation of average excess stock returns on U.S. stock
market than on CNAS stock market. As for the two additional factors, profitability factor
and investment factor are able to capture partially time-series variation of all three sets of
portfolios’ returns on U.S. stock market, while on Chinese stock market, the profitability
factor seems to be an explanatory factor only for the six Size-OP portfolios. Surprisingly,
we find that FF3F Model do not have cross-sectional explanatory power on U.S stock
market from July 2004 to May 2015.
65
Write between Chapter 1 and Chapter 2
Empirical research into the CAPM first documented that market risk was a factor that did
not perform well in explaining cross-sectional stock returns, then documented that other
factors - size factor SMB and value factor HML - did perform well in explaining stock
returns. Researchers had already observed that small stocks earned higher returns than large
stocks, and high book-to-market stocks earned higher returns than low book-to-market
stocks. Fama and French (1993) constructed factors on the basis of this observation, and
propose the FF3F Model. They demonstrated that the size and value factors did, indeed,
perform quite well in explaining cross-sectional stock returns. It is no exaggeration to say
that since the publication by Fama and French (1993), the FF3F Model has been accepted
as the most widely used expected returns model amongst researchers, and has become the
cornerstone of factor studies.
However, as is well-known that FF factors are based on purely empirical work, and there is
no clear theoretical foundation to identify the risk factors, which makes it one of the most
controversial asset pricing model. The huge success of FF3F Model lead to the question
that whether there is a body of theory to support the use of the particular factors that Fama
and French have identified. For next two decades a substantial body of literature devoted to
theoretical explanations for the explanatory power of SMB and HML.
There is debate amongst researchers who have attempted to explain why it is the two FF
factors explain stock returns and what risks are reflected in the size and book-to-market
factors. One dominant theoretical explanation is based upon the asset pricing theory already
established well before the empirical papers by FF – the Intertemporal CAPM (ICAPM) of
Merton (1973).
In the one-period CAPM, investors do not need to consider what happens outside of their
investment horizon. This is the basis for the intertemporal term in the model. While the
ICAPM assumes that investors trade continuously and maximize their expected utility of
lifetime consumption; investors care about what happens after the initial investment ends,
so will care about risks associated with future developments in the economy. So assets will
be priced, in a multi-factor model, according to investor expectations about future states of
the economy. For instance, investors might be concerned about future investment
opportunities. Over longer time periods, investment opportunities might shift as
expectations of risk change, resulting in situations in which investors may wish to hedge. In
other words, investors will seek to hedge against not only shocks to wealth as in the
traditional CAPM, but also against shocks to future investment opportunities.
The main difference between ICAPM and traditional CAPM is the additional state variables
such as economic variables that acknowledge the fact that investors hedge against changes
66
in the future investment opportunity set. The ICAMP is a linear factor model with wealth
and state variables that forecast changes in the distribution of future returns or income.
Our empirical results of chapter 1 demonstrate the applicability of FF3F Model on CNAS
stock market during July 2004 to May 2015; specifically, the size factor SMB explains the
cross-sectional variation of average excess stock returns well during the sample period.
Such results give rise to the questions: what are the economic explanation of FF factors on
Chinese stock market? Could they be explained in the context of ICAPM which has been
proved to be so by such as Petkova (2006) in the U.S stock market?
In order to answer the questions, we implement in the following chapter the theoretical
explanation based on time-varying investment opportunities in the frame of ICAPM using
data of CNAS stock market. Following Campbell (1996) who suggest using innovations in
state variables to forecast the changes in the future investment opportunity set, instead of
using the state variables directly for the empirical implementation of the ICAPM, we
consider the innovations of state variables that capture uncertainty about future investment
opportunities in our research; and examine whether FF factors proxy for innovations of
selected state variables on CNAS stock market.
67
2 Fama-French Factors and Innovations in
State Variables on Chinese A-Share Stock
Market
2.1 Economic explanation of Fama-French factors in the context of ICAPM ............ 69

2.2 Innovations in predictive variables and Vector Autoregressive approach ............ 81
2.3 Time-series evidence on Chinese A-share stock market ....................................... 85
2.4 Cross-sectional validation of five comparative models ....................................... 107
2.5 Conclusions .......................................................................................................... 113
This chapter examines whether the innovations of the four

predictive variables (aggregate dividend yield, one-month T-bill
rate, term spread and default spread) are able capturing excess
stock returns in both time-series and cross-section, further whether
Fama-French factors SMB and HML proxy for the innovations of
selected state variables that describe future investment
opportunities on Chinese A-share stock market. To derive the
innovation terms, we apply the vector autoregressive (VAR) method.
Following, Fama-MacBeth two-stage approach method is
implemented to perform the regressions for five models, and
Shanken correction is also performed to adjust for the Errors-in-
Variables problem in the cross-sectional regressions. Then the
comparisons are made among the five models.
69
Chapter 2 Fama-French Factors and Innovations in State Variables on Chinese A-Share Stock
Market
2.1 Economic explanation of Fama-French factors in the context of ICAPM
FF3F Model has achieved huge empirical success since it came up, which makes it a
popular as well as one of the most controversial asset pricing models. However, these
factors are based on purely empirical considerations, lack theoretical underpinnings, and
are built in a rather arbitrary manner. In particular, their economic links to systematic risk
are not clear. Therefore, the impressive performance of FF3F Model has aroused numerous
researches that trying to providing a clear economic interpretation of the factors HML and
SMB.
Some researchers attribute the success of FF factors to the survivor bias (Kothari et al.,
1995b), and data snooping (MacKinlay, 1995b). Lakonishok et al. (1994b) argue that the
B/M effect reflect investors’ incorrect inference of firms’ past earnings growth, suggesting
that investors undervalue firms with poor past performances while they overvalue firms that
have performed well in the past. Daniel and Titman (1997), however, suggest that it is the
characteristics of stocks rather than the covariance structure of returns that appear to
explain the cross-sectional variation in stock returns. While more researchers focus on the
alternative factors (such as macroeconomic variables) that are able to capture the variation
of equity returns in addition to the market return and have economic significance19.
The underlying economic links of FF factors are rather controversial, among plenty
competing explanations for the success of FF3F Model, following Petkova (2006), in this
chapter we focus on the one based on time-varying investment opportunities which is in the
context of Merton (1973a)’s Intertemporal Capital Asset Pricing Model (ICAPM hereafter).
2.1.1 ICAPM framework
Intertemporal Capital Asset Pricing Model (ICAMP) is put forward by Merton (1973a),
which as an improvement or extension of CAMP of Sharpe (1964) and Lintner (1965). It is
well documented in the literature that the CAPM is a static model which assumes betas
remain constant over time and that the return on the value-weighted portfolio of all stocks
is a proxy for the return on aggregate wealth. However, the single-period nature of CAPM
has been criticized, since most investors do not participate in financial markets for one year,
19
For example, Jagannathan and Wang (1996) include a labor income growth factor which performs
well in explaining the cross-section of average returns, Li, Vassalou, and Xing (2006) introduce
Sector Investment Growth Rates, Liew and Vassalou (2000) document that the size and B/M factors
predict future economic growth in some countries, Vassalou (2003) find that much of information
contained in the size and B/M factors is related to future GDP growth.
70
but instead for multiple years. Over longer time periods, investment opportunities might
shift as expectations of risk change, resulting in situations in which investors may wish to
hedge. For example, an increase in expected future returns will have a positive effect on
current consumption through decreased savings, in addition, an increase in the expected
volatility of returns will have a negative effect on current consumption through an increase
in precautionary savings (Khan, 2008).
The ICAPM assumes that investors trade continuously and maximize their expected utility
of lifetime consumption. It states that besides the market risk, the risk of unfavorable shifts
in the investment opportunity set, as approximated by the changes of the state variables,
will induce additional risk premiums and should be compensated 20 . In other words,
investors will seek to hedge against not only shocks to wealth as in the traditional CAPM,
but also against shocks to future investment opportunities. As specified by Dotsis (2015),
investors will bid up the prices of assets that do well when future investment opportunities
are expected to deteriorate, and consequently lower their expected returns. These assets
command a smaller risk premium because they increase the investor’s ability to hedge
against unfavorable changes in investment opportunities. On the other hand, investors will
require a higher premium for holding assets that do badly when future investment
opportunities worsen.
The main difference between ICAPM and traditional CAPM is the additional state variables
that acknowledge the fact that investors hedge against changes in the future investment
opportunity set. The ICAMP is a linear factor model with wealth and state variables that
forecast changes in the distribution of future returns or income (Cochrane, 2005).
In this study, the discrete-time version of the ICAPM is assumed to account for the cross-
sectional asset returns. Following Campbell (1996) who suggest using innovations in state
variables to forecast the changes in the future investment opportunity set, instead of using
the state variables directly for the empirical implementation of the ICAPM, we consider the
innovations of state variables that capture uncertainty about future investment opportunities
in this research. According to ICAPM, both the excess market return and innovations in
state variables that forecast changes in the future investment opportunity set should show
up as pricing factors in the cross-section of asset returns.
We assume the unconditional expected returns can be expressed in the frame of ICAPM as
follows:
20
Campbell (1993) extends Merton (1973)’s model to a discrete-time ICAPM and derives a simple
non-consumption based model with two risk factors: the unexpected current period return on the
market portfolio, and news about future expected returns on the market portfolio.
71
Market
 
K
E  Ri ,t   R f ,t  i ,m   m ,t     i , k   k ,t , i (2.1)
k 1
where E  Ri ,t   R f ,t is the excess return of asset i, R f ,t is the risk-free rate of return,  m ,t is

the market risk premium,   k ,t is the risk price for innovations of state variable k. The  s
are the coefficients that can be estimated through following TSR:
Ri ,t  R f ,t   i ,t   i ,m  RM ,t  R f ,t     i , k  tk   ei ,t
K
(2.2)
1
where Ri ,t  R f ,t is the excess return of asset i at time t, RM ,t  R f ,t is the excess return of the
market portfolio at t, and tk is the innovation of state variable k at time t.
The beta terms in equation (2.1) measure how much risk should be priced while the gamma
terms measure the price of risk, with the excess market return and innovations of state
variables as risk factors.
2.1.2 What are the proper candidates for the state variables?
The ICAPM suggests that we should use variables that forecast stock market returns as
proxies for investment opportunities, however, it provides little guidance for identifying
them. Thus what are the proper candidates for the state variables has raised many
discussions. Theoretically, state variables should be factors that have predicting power of
the future investment opportunity set. Empirically, numerous researchers have explored
various candidates that for the most part, are macroeconomic variables that related to
business cycle fluctuations.
Fama (1981), Fama (1990), Geske and Roll (1983), and Schwert (1990) document that U.S.
stock returns are positively related to the future growth rate in the gross national product
(GNP)21. Stock and Watson (1989) find interest rates are particular useful predictors of
future economic activity. Later, Brennan et al. (2004) develop and test a model assumes
that the investment opportunity set is completely described by real interest rate and the
maximum Sharpe ratio. The estimated real interest rate and Sharpe ratio both show strong
business cycle-related variation and both state variables have significant risk premium in
the cross-sectional asset test. Liew and Vassalou (2000) suggest to use the return on a
21
Mullins and Wadhwani (1989) report a similar relationship in Germany and the United Kingdom.
72
market portfolio, dividend yield, short-term interest rates, term spreads, growth in the Gross
Domestic Product (GDP) and the industrial production as indicators for the business cycle.
Other studies that identify significant predictors of the equity risk premium include: Lintner
(1975), the interest rate; Campbell and Shiller (1988) and Fama and French (1988), the
market dividend yield; Fama and French (1989), the term spread and the junk bond yield
spread; Kothari and Shanken (1997), the book-to-market ratio; Dumas and Solnik (1995)
exchange risk.
Among the attempts to find the economic variables that are able to help explain expected
asset returns, Fama and Schwert (1977) prove that common stock returns associated with
expected and unexpected components of the inflation rate22; Jagannathan and Wang (1996)
include the return on human capital (using labor income growth as a proxy) as a new factor
in addition to market beta; Campbell (1996) also find the future labor income growth a
significant priced factors in determining excess stock returns; while Parker and Julliard
(2005) and Hansen et al. (2005) examine the relationship between expected asset returns
and the future consumption.
Chen et al. (1986) explore series of economic state variables related to industrial production,
inflation (change of Consumer Price Index), risk premium (the spread between high- and
low-grade bonds), term structure (the spread between long and short interest rate),
consumption, etc., and examine whether innovations in those macroeconomic variables
help explain the cross-section of average returns on NYSE stocks. They find several of
these economic variables were found to be significant in explaining expected stock returns,
variables related to industrial production, risk premium, term structure, and, somewhat
more weakly, variable related to inflation.
Similarly, Keim and Stambaugh (1986) show that default spread and term spread forecast
stock and bond returns. Moreover, Chen (1991) also confirm that the default spread, the
term spread, the one-month T-bill rate, the lagged industrial production growth rate, and the
dividend-price ratio are important determinants of future stock market returns. However, Li
(1997) implement tests of ICAPM using forecasting variables (a yield spread measuring
term premium, a yield spread for default risk, the dividend yield, one-month T-bill rate and
the lagged market return) as risk factors; the evidence rejects the hypothesis that the
forecasting variables are the risk factors that explain the time-series and cross-sectional
variation in expected stock returns.
22
See also researches on inflation: Shi et al. (2015), Adams et al. (1999), Bottazzi and Corradi (1991)
on Italy market.
73
Market
Ferson and Harvey (1991) strengthen the evidence that the predicted variation of asset
returns is related to their sensitivity to the economic variables through the analysis of a
group of six economic variables. Results turn out that the premiums associated with interest
rate, term structure shifts, and default spreads are the important variables for capturing the
predictable variation of asset returns in addition to the market risk premium.
Furthermore, Campbell (1987), Campbell (1991), Campbell and Shiller (1988a), Fama and
French (1988), Fama and French (1989), Harvey (1989) and Kothari and Shanken (1997),
have proposed various candidates related to the yield curve shape and aggregate dividends.
For instance, Campbell (1987) document that the term structure of interest rates predicts
excess stock returns in U.S.; Campbell (1991) shows that the unexpected stock returns are
related to changes in expected future dividends or expected future returns. Kothari and
Shanken (1997) provide evidence that B/M ratio and dividend yield are related to time-
series variation in expected stock returns during the period 1926 to 1991. 23 Fama and
French (1989) use three variables to forecast returns, which are dividend yield on the value-
weighted NYSE portfolio, the default spread defined as the difference between yield on a
portfolio of 100 corporate bonds and yield on a portfolio of bonds with Moody’s Aaa
ratings, and the term spread between the yield on the Aaa corporate bond portfolio and the
one-month T-bill rate.
Inspiring by those findings and as Fama and French (1993, 1995, 1996) had themselves
suggested that their SMB and HML factors could be interpreted as proxies for state
variables which describe time variation in the future investment opportunity set in the
context of Merton’s ICAPM. 24 Furthermore, one of the macroeconomic explanations
behind the success of FF3F Model is based on the time-varying investment opportunities,
FF two factors SMB and HML are proxy for state variables. Many types of research are
proceeded by relating the FF factors to macroeconomic variables that are closely related to
business cycle fluctuations. Numerous empirical evidence in the literature suggests that the
size factor SMB and value factor HML indeed carry information about future investment
opportunities.
For choosing variables that have forecasting power for future investment opportunities,
Fama (1991), Campbell (1996) have pointed out that ICAPM should not be used as only
23
Many studies have found dividend yield to have predictive power in both cross-section
Litzenberger and Ramaswamy (1979) and time series Rozeff (1984), Fama and French (1988),
Fama and French (1989), and Campbell and Shiller (1988b).
24
Simpson and Ramchander (2008) provide evidence that the FF3 model outperforms the standard
CAPM in its ability to capture surprises related to various macroeconomic indicators.
74
criteria for selecting factors. Only factors that forecast future investment opportunities
should be included in the model.
Both Liew and Vassalou (2000) and Vassalou (2003) report that FF factors SMB and HML
convey significant information about future GDP growth not present in the market portfolio.
Especially, Liew and Vassalou (2000) provide evidence by using data from ten developed
countries. Vassalou (2003) find that when news related to future GDP growth presented in
the asset pricing model, SMB and HML lose much of their explanatory power in the cross-
sectional variation of average asset returns. Hanhardt and Ansotegui Olcoz (2008) found
that this result extends to twelve countries of the Eurozone.
Prior study of Chan et al. (1985) explore multifactor pricing equation that consists of five
economic variables to explain the firm size effect. The five variables are (1) the change in
the state of the economy measured by the growth rate of industrial production; (2) the
change in expected inflation; (3) the difference between the realized inflation rate; (4) the
change in the long term rate measured by the difference between the return of a portfolio of
long-term government bonds and the T-bill rate; (5) the changing risk premium measured
by the behavior of bonds of different perceived riskiness. Among those economic variables,
a measure of the changing risk premium and a measure of the changing state of the
economy explained a large portion of the size effect.
Xing (2008b) claims that FF value factor HML may approximate by an investment growth
factor defined as the difference in returns between low-investment stocks and high-
investment stocks. Campbell and Vuolteenaho (2004), Brennan et al. (2004) and Petkova
and Zhang (2005) all show that the value premium is correlated with innovations in their
measures of investment opportunities.
Fama and French (1993) examine two bond-market factors TERM (the difference between
the long-term government bond return and one-month T-bill rate) and DEF (the difference
between the return on a market portfolio of long-term corporate bonds and the long-term
government bond return) as measures of unexpected changes in interest rates and the
likelihood of default separately. They confirm that the tracks of both factors show up
clearly in the time-series variation of stock returns. They also find both factors capture
common variation in stock and bond returns when they are examined alone in the bond
market or the stock market.
Both term spread and default spread are well known to forecast aggregate stock market
returns (Fama and French (1989), Fama and French (1993) and Keim and Stambaugh
(1986)). Term spreads and default spreads have been shown to have time series predictive
ability to forecast stock market returns (Fama and French (1988), Stock and Watson
75
Market
(1989)), so they are also widely used as potential conditioning variables in cross-sectional
tests.
Hahn and Lee (2001, 2006) emphasize in their articles that since term spread and default
spread are two of the most widely used proxies for future interest rates and time-varying
risk premium separately, they are likely to capture well the hedging concerns to investors
associated with variations in interest rates and risk premium. The authors choose both term
spread and default spread as proxies for SMB and HML and specify a three-factor model in
which the factors are the excess market return, changes in the default spread (def) and
changes in the term spread (term). They conclude that def and term capture most of the
systematic risks proxied by the FF factors SMB and HML in both time-series and cross-
sectional dimensions (def is proxy for size factor while term proxy for B/M factor), and
in the presence of def and term, FF factors are superfluous in explaining the size and
B/M effects.
Following Campbell’s (1996) argument that instead of relying on choosing important

macroeconomic variables as empirical implementations of ICAPM, the factors that are
related to innovations in state variables which forecast future investment opportunities
should be considered in the model. Researchers like Campbell and Vuolteenaho (2004),
Petkova (2006), In and Kim (2007) and Aretz et al. (2010) implement their research with
innovations of state variables rather the economic variables themselves.
Petkova (2006) chooses a set of four innovations of state variables, including the short-term
T-bill rate, term spread, aggregate dividend yield and default spread instead of some
important macroeconomic variables. And she comes to a conclusion that FF factors SMB
and HML are significantly correlated with innovations in state variables that describe
investment opportunities and a model which uses innovations in SMB and HML and in the
predictive variables explains the cross-sectional average returns better than the FF3F model.
As a result, the author concludes that Fama-French factors proxy for innovations in
predictive variables. More specifically, she denotes that HML proxies for a term spread
surprise factor in returns, while SMB proxies for a default spread surprise factor.
Using the same set of four state variables as Petkova (2006), In and Kim (2007) adopt a
new approach of wavelet analysis and examine to what extent FF factors SMB and HML
share information with the innovations of state variables that describe investment
opportunities. It is found that SMB and HML seem to play only a limited role in capturing
alternative investment opportunities in the short run, but they share much information with
alternative investment opportunities in the long run.
76
More recently, Aretz et al. (2010) consider a series of innovations of state variables that
include innovations in economic growth expectations, inflation, the aggregate survival
probability, the term structure of interest rates, and the exchange rate. They prove that most
of the macroeconomic factors considered are priced, and B/M ratio, size and momentum
capture cross-sectional variation in exposures to those macroeconomic factors. Specifically,
the authors show B/M conveys useful information about term structure risk and economic
growth, and size conveys information about default risk and term structure risk.
However, there are researchers deduce different conclusions, the ability of macroeconomic
state variables to predict portfolio excess returns and the significance of factor loadings in
TSRs are not encouraging. For instance, Campbell’s (1996) own results are mixed. In his
paper, he argued that CAPM ignores time variation in expected stock returns and the fact
that human capital is also an important component of wealth, but still a good approximate
model of stock and bond pricing in some limited senses.
Chen (2002) develop a model with time-varying expected market returns and time-varying
market volatilities to reflect the changes in the investment opportunity set. The author
examines the size effect, value effect and momentum effect but neither the value effect nor
the momentum effect can be explained using changes in the investment opportunity set.
Thus the author concludes that accounting for the changes in the investment opportunity set
does little in explaining the cross-section of stock returns.
Shanken and Weinstein (2006) and Lewellen et al. (2010) express rather pessimistic views.
The former re-examine the five macroeconomic factors studied by Chen et al. (1986) and
Chan et al. (1985). Contrary to the previous literature, the authors show that only the
industrial production factor is significantly priced in the overall period 1958 to 1983; the
bond return premium, a highly significant factor in the earlier studies, is insignificantly
negative for the whole period. The five factors only account for about 25% or 30% of the
time-series variation in 20 size portfolio returns and the authors fail to find the evidence of
factor pricing over the sub-period 1968 to 1977. In the latter article, the authors review the
models suggest new risk factors to help explain expected returns (growth in
macroeconomic output and investment and innovations in state variables included), which
seem to do a good job explaining the size and B/M effects. They critique the empirical
methods used in the literature and offer improving empirical tests and find that several
models do not work as well as originally advertised.
Especially, on the contrary to Petkova (2006), Lioui and Poncet (2011) find that the method
that used to make innovations in predictive variables is rather arbitrary, and loadings on
innovations of state variables are rarely significant when tested on the 25-portfolios sorted
by size and B/M equity. These innovations of state variables have extremely limited
77
Market
explanatory power in both time-series and cross-section regressions when extending the
portfolio universe to include 30 industry portfolios. They also concluded that there is no
proof that the FF factors proxy for time-varying investment opportunities, and the
explanatory power of innovations on SMB and HML is almost inexistent in time-series and
marginal in cross-sections, while SMB and HML themselves remain as significant as in the
FF3F Model.
More recently, Boons (2016) find that a series state variables such as the default spread, the
term spread, the short-term T-bill rate, robustly forecast macroeconomic activity. However,
they find that FF factors SMB and HML are unable to drive out the risk premiums of those
state variables and their underlying characteristics are able to do so only partially.
2.1.3 Evidence from outside of the U.S. market
Though the conflict point of views on whether macroeconomic variables are able to explain
expected stock returns, the evidence that stock prices tend to fluctuate with economic news
is supported by numerous empirical literatures. For instance, the early studies of Fama
(1981), Fama (1990), Chen et al. (1986), Schwert (1990), and Ferson and Harvey (1991)
have found that macroeconomic variables have explanatory power for U.S. stock returns.
While this observation is not found only in U.S., such as Asprem (1989), Beckers et al.
(1992), Ferson and Harvey (1993) and Cheung et al. (1997). have reached a similar
conclusions using data of other international market out of U.S.
Asprem (1989) investigates the relations between stock indices, asset portfolios and
macroeconomic variables in ten European countries. The results show that changes in stock
prices are correlated to some measures of real economic activity, in particular positive
related to future industrial production, exports, the yield curve in the U.S.; and negatively
related to employment, the exchange rate, imports, inflation and interest rates. The author
suggests these economic variables may be representatives of state variables in the ICAPM.
Hanhardt and Ansotegui Olcoz (2008) discuss the economic rationale of FF factors SMB
and HML in Eurozone market in the context of Merton’s (1973) ICAPM. The authors also
extend their research by including a momentum factor of Carhart (1997) WML, and they
test to what extent the profitability of the factors can be related to future economic growth
measured by growth in the gross domestic product (GDP) in the Eurozone. Their results
document that only the size factor SMB seems to contain strong and robust information
with respect to future growth in GDP, while they fail to find the same significance for HML
and WML (there appears to be a positive but not significant relationship between HML and
GDP growth,). Thus they conclude that at least SMB, and to some extent HML, may serve
78
as state variable that predict future changes in the investment opportunity set in the context
of ICAPM.
Docherty et al. (2013) provide an empirical analysis of whether empirical regularities (they
denote the size, value and momentum premium as empirical regularities) previously
identified in the Australian market are related to macroeconomic risk factors such that they
may be considered as the state variables of Merton’s (1973) ICAPM. The authors examine
two groups of macroeconomic fundamentals in their paper. The first group are the state
variables employed by Chen et al. (1986) 25 , and the second group are a series of
macroeconomic forecast variables shown in the prior literature to predict equity returns.
They report that all three empirical regularities examined change with the business cycle
(HML is negatively related to the business cycle while SMB and momentum factor has the
positive direction). Their results support the contention that the empirical regularities may
be explained as macroeconomic risk factors in the Australian stock market. The authors
suggest that the three anomalies are correlated with innovations of those macroeconomic
variables that describe future investment opportunity set.
Cheung et al. (1997) examine two potential sources of international real return variation:
changes in expected future cash flows and changes in discount rates in 18 national stock
markets, and the authors give evidence that the global economic variables that proxy for the
two sources of return variation significantly capture large fraction of total variation of the
international stock market returns.
Cheung and Ng (1998) investigate the relationship between stock market indexes and
measures of aggregate economic variables (the real oil price, real output, real money supply,
and real consumption) on five countries stock market (Canada, Germany, Italy, Japan, and
the U.S.) by adopting the cointegration approach. The authors find evidence of long-run
comovements between the returns on national stock indexes and measures of the country-
specific aggregate economic real variables. They further explore an error correction model
which provides incremental information about the variation of stock returns that not found
in other measures of return variation such as dividend yields, default and term spreads, and
future GNP growth rates.
25
The authors examine all the state variables documented by Chen et al. (1986) except the default
spread: unexpected inflation, changes in expected inflation, industrial production growth and the
term spread. As the authors claim that the default risk spread is not included in their forecast model
due to the illiquidity of Australian bond markets.
79
Market
Charles et al. (2016) investigate the relationship between stock returns and a range of
economic fundamentals including short-term interest rates, several financial ratios
(dividend-price ratio, dividend yield, E/P ratio, dividend-payout ratio) and technical
indicators (price pressure, change in volume) on international markets (16 Asia-Pacific and
21 European stock markets). They show that the financial ratios have the weak predictive
ability for stock returns while the price pressure and the short-term interest rate appear to
have strong predictive power for stock return.
In emerging market, Hosseini et al. (2011) examine the relationship between stock market
indices and four macroeconomic variables (crude oil price, money supply, industrial
production and inflation rate) in China and India. Their results indicate that there exists the
link between four macroeconomic variables and stock market index in both long and short
run in both countries.
Pramod Kumar and Puja (2012), Tripathi and Seth (2014) and Gaur et al. (2015) investigate
the impact of selected macroeconomic variables on the performance of Indian stock market,
all of them find significant relationship between stock price and the selected
macroeconomic factors (such as industrial production, inflation, short-term interest rate,
exchange rate). On Malaysia stock market, Siti Noorahayusolah (2011) find a series of
macroeconomic factors (inflation, interest rates, money supply, and exchange rates) has a
significant impact on stock market returns.
Liang (2013) and Liang and Willett (2015) provide evidence that the performance of
Chinese stock market is related to both its domestic economic fundamentals, the policy-
driven factors such as exchange rate and bank deposits and bank loans have strong impacts
on stock performance. However, the real economic factors such as industrial production
seem not as significant as the policy-driven factors in explaining Chinese stock returns.
The latest empirical study of Mu (2016) investigates the relationship between seven
selected macroeconomic variables (interest rate, inflation, oil price, unemployment rate,
industrial production index, money supply, and exchange rate) and the stock markets in the
US, Germany, and Hong Kong. Both short-term and long-term relationships between
macroeconomic variables and the stock markets are shown in all the three countries.
Plenty researchers confirm the ability of FF factors (at least the size factor SMB) in
explaining average excess stock returns on Chinese stock market. We consider several
special features of Chinese stock market, and examine FF3F Model using data on CNAS
stock market. We provide evidence that FF3F Model explains time-series variation of
average excess stock returns well, and there always exists size premium. Though there are
researchers who find the relevance between the performance of Chinese stock market and
the domestic economic variables, rare have examined whether the economic underpinning
80
of FF factors are related to the economic fundamentals of Chinese stock market, which is
what we are going to investigate in this study.
Following the framework of Petkova (2006), we choose the same set of four variables in
this study  including aggregate dividend yield, short-term T-bill rate, term spread and
default spread. Our choice is also consistent with the standpoint of Campbell (1996) that
instead of choosing those important macroeconomic variables, variables which have
forecasting power for future investment opportunities should be considered in the model.
The four variables are choosen to relate two aspects of investment opportunities, the yield
curve and the conditional distribution of asset returns, and all these variables have been
frequently used among literature.
We apply the vector autoregressive (VAR) approach of Campbell (1996) in this study to
obtain the innovation terms of selected state variables and examine the performance of
innovations of predictive variables on CNAS Stock Market during December 2006 to May
2015. In addition, we also investigate whether innovations of selected variables have the
proxy ability of FF factors in China.
This chapter proceeds as follows: Section 2 presents our data and the VAR process to
extract innovations of state variables; the TSR results of the five comparing models are
demonstrated in section 3; section 4 test the cross-sectional validation of the five models;
finally, the conclusions and discussions are in the last section.
2.2 Innovations in predictive variables and Vector Autoregressive approach
2.2.1 State variables
We choose the same set of state variables as Petkova (2006) in addition to the FF factors in
our empirical test: dividend yield (DIV), term spread (TERM), default spread (DEF) and
one-month T-bill rate (RF), which are among the most common used economic variables in
the literatures as illustrated previously. The four predictive variables are chosen to model
two important aspects of the investment opportunity set, the yield curve and the conditional
distribution of asset returns. “The ICAPM dictates that the yield curve is an important part
of investment opportunity set” Petkova (2006), Litterman and Scheinkman (1991) indicate
that the two most important factors driving the term structure of interest rates are its level
and its slope. One-month T-bill rate and the term spread are to capture variations in the
level and slope of the yield curve. Petkova (2006) points out that the conditional
81
Market
distribution of asset returns is also a crucial aspect of investment opportunity set in the
context of ICAPM. And growing literature indicates that the conditional distribution of
asset returns which are characterized by its mean and variance, varies over time. The
aggregate dividend yield, the default spread, and interest rates are among the most
generally identified variables by numerous literature. Table 2.1 lists a partial of papers that
document the relationship between the performance of equity market and the four state
variables that we are going to apply in the following research.
Table 2.1 List of papers that document time variation of excess asset return and the state
variables they use
State Variables Paper/ author
Campbell and Shiller (1988a), Fama and French (1988), Fama and
Dividend yield
French (1989), Kothari and Shanken (1997), Chen (1991)
Campbell (1987), Keim and Stambaugh (1986), Fama and French

Term spread
(1989), Chen et al. (1986), Chan, Chen, and Hsieh (1985), Chen (1991)
Fama and French (1989), Chen et al. (1986), Chan, Chen, and Hsieh
Default spread
(1985), Chen (1991)
Brennan et al. (2004), Fama and Schwert (1977), Stock and Watson
Short-term T-bill rate
(1989), Chen (1991)
In addition to FF factors SMB and HML, the four other state variables we selected on
Chinese market are:
Aggregate Dividend yield (DIV) is the sum of dividends over the last 12 months, divided
by the actual value of the market index. Our study focuses on CNAS Stock Market, so we
take all the CNAS stocks (both Shanghai A-share Stock Market Index and Shenzhen A-
share Stock Market Index) present value as the actual value of the market index.
Term spread (TERM) is the difference between the yields of a 10-year government bond
and a 1-year government bond. We download Generic China 10Y Government Bond and
Generic China 1Y Government Bond which fit the definition.
Default spread (DEF) is the difference between the yields of a long-term corporate Baa-
rated bond and a long-term government bond. Mention that the Baa-rated bond is rated by
Moody's rating system, which is not available in Chinese bond market through Bloomberg.
Thus we compute the default spread as the difference between the yields of a long-term
82
corporate bond index (China 10Y Corporate Bond) and a long-term government bond index
(China 10Y Government Bond).
Short-term Treasury bill rate (RF) is generally the one-month Treasury bill rate or called
risk-free rate which is a typically proxy for the return on a one-month Treasury bill. In
China, we use one-month fixed deposit rate as the proxy for the risk-free interest rate.
Monthly value-weighted excess market returns (CNAS stock market, including both
Shanghai A-share stock Market and Shenzhen A-share Stock Market), one-month deposit
rate, aggregate dividend yields, and yields of government bonds and corporate bonds are
obtained from Bloomberg from November 2006 to May 2015. Starting from November
2006 rather than 2004 (which is the starting year in Chapter 1), because the available data
of 'Term Spread’ from Bloomberg begins from November 2006. FF factors SMB and HML
are constructed the same way using the six Size-B/P portfolios in Chapter 1.
2.2.2 Vector Autoregressive method
According to ICAPM, only the unexpected component of the state variable should
command a risk premium. The unexpected component is normally we called innovations
(or unexpected shocks to state variables). Instead of using the state variables themselves
directly for the empirical implementation of the ICAPM, Campbell (1996) suggests using
innovations in such state variables to forecast the changes in the future investment
opportunity set.
To derive the innovations of state variables26, in the article of Chen et al. (1986), the author
proposes using a vector autoregressive model to derive the residuals of the state variables as
the unanticipated innovations in the economic factors. Campbell (1991) also state that “The
resulting vector autoregressive (VAR) system can be used to calculate the impact that an
innovation in the expected return will have on the stock price, holding expected future
dividends constant” Following Campbell (1996), Petkova (2006) also adopts the first-order
vector auto-regression model to derive the innovations of the state variables in her model.
All the studies observed significant risk premiums induced by innovations in the state
variables.
The Vector Autoregressive (VAR) Model generalizes the univariate autoregressive model
(AR model) by allowing for more than one evolving variable. What is more, the VAR
26
There are other approaches to estimate the innovations of state variables: Brennan et al. (2004)
assumed the Ornstein–Uhlenbeck process whereas Hahn and Lee (2006) use the simple changes of
state variables.
83
Market
model is one of the most successful, and easy to use models for the analysis of multivariate
time series. As Zivot and Wang (2007) say that “the VAR model has proven to be especially
useful for describing the dynamic behavior of economic and financial time series and for
forecasting”.
Following Petkova (2006), we adopt the VAR approach proposed by Chen et al. (1986) and
Campbell (1991) in this study. All variables in a VAR enter the model in the same way:
each variable has an equation explaining its innovation based on its own lags and the lags
of the other model variables. We write the excess market return as the first element of the
vector zt , the other elements are the state variables which proxy for changes in the
investment opportunity set. The assumption is that the vector 𝑧𝑡 follows a first-order VAR:
zt  Azt 1  ut (2.3)
where, zt is a k ×1 vector which has k time-series variables, and zt 1 is called the 1- lag of
zt , which has the one-period back observation of the variables in zt . A is a k × k matrix
which is known as the companion matrix of the VAR. The error term ut is also a k ×1
vector, and the residuals in ut are the innovation terms that are regarded as the risk factors.
Petkova (2006) underlines that “these innovations are risk factors since they represent the
surprise components of the state variables that proxy for changes in the investment
opportunity set”.
The assumption that the VAR is first-order is not restrictive since a higher-order VAR can
always be stacked into first-order (companion) form in the manner discussed by Campbell
and Shiller (1988a).
2.2.3 Innovations in state variables
In our research, we define the first element of vector zt (equation (2.3)) is the excess
market return RM ,t  R f , denoted as Rm ,t , and the other elements following are dividend
yield ( DIVt ), term spread ( TERM t ), default spread ( DEFt ), one-month deposit rate or
risk-free rate( RFt ) and two FF factors ( SMBt and HMLt ). Thus we have a set of seven
variables, consistent with (Campbell, 1991) and (Petkova, 2006), for simplicity, all
variables in the vector zt have zero means or have been demeaned. The first-order VAR is:
84
 Rm,t   Rm,t 1 
 DIV   DIV 
 t
  t 1

 TERM t TERM t 1 
   
 DEFt  = A  DEFt 1  + ut (2.4)
 RFt   RFt 1 
   
 SMBt   SMBt 1 
 HMLt   HMLt 1 
Where A is a 7×7 matrix, and ut represents a 7×1 vector of innovations for each element in
the state vector zt . From ut we extract six innovation series corresponding to the dividend
yield, one-month T-bill rate, term spread, default spread, SMB and HML, which are
denoted as IDIV, IRF, ITERM, IDEF, ISMB and IHML.
Campbell (1996) emphasizes that it is hard to interpret estimation results for a VAR factor
model unless the factors are orthogonalized and scaled in some way. In his research, he
orthogonal the innovations of the state variables to both excess market return and labor
income. In Petkova’s (2006) article, the author implements the similar way as Campbell:
the residuals ut are orthogonalized from the market return Rm ,t , and scaled to have the
same variance as the market return. Aretz, Bartram, and Pope (2010) orthogonalize the
market return with respect to their other macroeconomic fundamentals in order to isolate
the variation in market returns not attributable to the macroeconomic fundamentals.
Following Petkova, in this study, the innovations of state variables are orthogonalized to
the market factor separately. Precisely, the innovation of excess market return remains
unchanged, innovation of dividend yield IDIV is orthogonalized to the excess market return;
similarly, then innovation of the one-month T-bill rate IRF is orthogonalized to market
factor. We do in this way to the other innovations of state variables ITERM, IDEF, ISMB
and IHML. The innovations of the state variables are not orthogonalized from each other
because this could add noise through the arbitrary ordering of the variables Boons (2016).
2.3 Time-series evidence on Chinese A-share stock market
This section is to examine the performance of the innovations of state variables which
proxy for the future investment opportunities on CNAS stock market, and whether FF
factors are proxies for innovations of these predictive variables. And we make comparisons
85
Market
among models that contain different factors, involving TSRs on excess market return,
innovations of four state variables and innovations of FF factors (Model 1); regressions on
excess market return, innovations of four state variables and original FF factors (Model 2);
regressions on excess market return and innovations of the four state variables (Model 3);
regressions only on excess market return and innovations of FF factors IHML and ISMB
(Model 4); and time-times regressions on the original FF3F Model (Model 5).
2.3.1 Fama-French factors and innovations of state variables
2.3.1.1 Statistics description
Table 2.2 represents the summary statistics of the original FF three factors, the four state
variables, and their innovations during the period December 2006 to May 2015 (102
months) in China. The mean of innovations of state variables (IDIV, ITERM, IDEF, and
IRF) and innovations of FF factors (ISMB and IHML) are all close to zero. Not surprising
that the t- statistic of SMB is significant at 5% confidence level, however, neither the
Table 2.2 Summary statistics of FF factors, state variables and their innovations (December
2006-May 2015, 102 months)
This table presents the summary statistics of original FF three factors, the four state variables and
their innovations. Sd error is the standard error, and S.D. is the standard deviation. Rm ,t is the excess
market return.
Mean Sd error t-stats Median S.D. Variance Kurtosis Skewness
Rm ,t -0.0025 0.0094 -0.2662 0.0050 0.0945 0.0089 0.8995 -0.6524
SMB 0.0138 0.0040 3.4643 0.0173 0.0402 0.0016 2.8062 -0.8652
HML -0.0024 0.0035 -0.6823 -0.0058 0.0352 0.0012 3.4058 0.5215
DIV 0.0153 0.0006 24.8090 0.0150 0.0062 0.0000 -1.0382 -0.1687
TERM 0.0101 0.0006 18.2947 0.0088 0.0056 0.0000 -0.7151 0.6089
DEF 0.0141 0.0003 43.9995 0.0145 0.0032 0.0000 1.7629 -0.9073
RF 0.0130 0.0005 28.0855 0.0136 0.0047 0.0000 -0.4566 -0.1460
IDIV 0.0000 0.0001 0.0000 -0.0001 0.0008 0.0000 4.0478 -0.1081
ITERM 0.0000 0.0002 0.0000 -0.0001 0.0022 0.0000 1.3729 0.4382
IDEF 0.0000 0.0001 0.0000 0.0000 0.0013 0.0000 0.6156 -0.1258
IRF 0.0000 0.0002 0.0000 -0.0003 0.0020 0.0000 2.8160 0.9862
ISMB 0.0000 0.0040 0.0000 0.0031 0.0399 0.0016 2.7858 -0.7851
IHML 0.0000 0.0034 0.0000 -0.0016 0.0342 0.0012 3.3387 0.3595
86
innovations of state variables nor the innovations of FF factors are significant. Meanwhile,
all the four state variables have high t-value. The significant t-stats indicate the high
possibility to be priced in equity returns.
Primarily, we perform regressions with the single innovation in addition to excess market
return (the results are presented in Appendix D), in order to have the first glance of whether
such small values of innovations related to the average excess stock returns. It is shown that
among the four innovations of state variables, 23 out of 25 loadings on IDIV is significant
at 5% confidence level and negatively related to stock returns; while none of the loadings
on ITERM and none of the loadings on IDEF is significant or exhibit any significant
systematic patterns related to size or B/P ratio. There are 11 out of 25 loadings on IRF that
are significant, and it seems that the loadings are not related to size or B/P neither. It is
probably that IDIV and IRF are able to capture variation in average excess stock returns,
while the ITERM and IDEF probably not. Whether innovations are related to size or B/P
ratio is not clear.
Table 2.3 shows the correlation matrix for all the risk factors which are demeaned FF three
factors ( Rm ,t , SMB and HML), the innovations of state variables (IDIV, ITERM, IDEF,
and IRF) and the innovations of SMB and HML (ISMB and IHML). Notably, SMB and
HML are very highly correlated with their innovations with correlation coefficients 0.9943
between SMB and ISMB, 0.9720 between HML and IHML. Similarly, Petkova (2006) find
the returns on FF factors are also very highly correlated with their respective innovations
(the correlation
Table 2.3 Correlation coefficients among risk factors
This table presents the correlation coefficients among risk factors (FF three factors, innovations of
state variables, and innovations of SMB and HML). FF three factors are all demeaned. The
underlined numbers show the most highly correlated coefficients.
Rm ,t SMB HML IDIV ITERM IDEF IRF ISMB IHML
Rm ,t 1
SMB 0.0065 1
HML 0.1926 -0.3495 1
IDIV 0.0000 -0.2593 0.0360 1
ITERM 0.0000 0.1557 -0.3077 0.0337 1
IDEF 0.0000 0.0184 0.0424 -0.1224 -0.2755 1
IRF 0.0000 -0.2288 0.2380 0.0686 -0.4842 0.1131 1
ISMB 0.0000 0.9943 -0.3533 -0.2608 0.1566 0.0185 -0.2301 1
IHML 0.0000 -0.3614 0.9720 0.0370 -0.3166 0.0436 0.2449 -0.3635 1
87
Market
is 0.92 between SMB and ISMB, 0.90 between HML and IHML) in U.S., thus she suggests
that “the returns on the HML and SMB portfolios are good proxies for the innovations
associated with these variables”. ITERM and IRF are also significantly correlated with the
correlation coefficient -0.4842. The high correlation coefficients indicate the factors
probably share common information in explaining equity returns.
Considering the absolute value, the correlation among other risk factors are relatively weak,
still there are factors share common information to some extent, such as SMB and HML (-
0.3495), SMB and IDIV (-0.2593), SMB and IRF (-0.2288), HML and ITERM (-0.3077),
HML and IRF (0.2380). The correlation coefficients are almost close to zero between the
excess market return Rm ,t and the innovations, from which we can infer that the
innovations have no correlation to the market factor.
2.3.1.2 Relation between FF factors and innovations of state variables
To test whether Fama-French factors are proxies for innovations of state variables on
CNAS stock market. First of all, we perform two types of TSRs: (1) regressions of SMB
and HML respectively on market factor and innovations of state variables, (2) regressions
of each innovation of the state variables on FF three factors. Results of the two types of
TSRs are presented separately in Table 2.4 and Table 2.5, showing intuitively the relation
among those risk factors.
We first examine the relationship between two FF factors (SMB and HML) and four
innovations of state variables estimated from a VAR process, controlling for the market
factor, using the following two regression equations:
SMB  ai  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF

(2.5)
 i ,IRF IRF  i ,t
HML  ai  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF

(2.6)
 i ,IRF IRF  i ,t
Table 2.4 reports the regression coefficients and corresponding t-stats of equation (2.5) and
(2.6). Corresponding t-stats are in the parentheses below the regression coefficients,
corrected for heteroscedasticity and autocorrelation using the Newey-West estimator with
five lags.
88
Table 2.4 Estimated coefficients of SMB and HML from risk factor regressions
This table presents the regression results of two FF factors, SMB and HML, on the other risk factors
(market factor and innovations of state variables). The adjusted R-square is represented in the last
column. Three FF factors and four state variables are all demeaned, and the t-statistics are corrected
for heteroscedasticity and autocorrelation using the Newey-West estimator with five lags. The
sample period is from December 2006 to May 2015.
Regressions:
SMB  ai  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF  i ,IRF IRF  i ,t
HML  ai  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF  i ,IRF IRF  i ,t
ai  i ,m i ,IDIV i ,ITERM  i ,IDEF  i ,IRF Adj. R 2
SMB 0.0012 0.0028 -12.1704 1.6072 1.0019 -3.4078 0.0720

(t-stats) (0.4096) (0.0622) (-3.3173) (0.6044) (0.3028) (-1.6984)
HML 0.0000 0.0717 1.4015 -4.1733 -1.0707 1.9352

0.1004
(t-stats) (0.0074) (1.4730) (0.4380) (-2.0204) (-0.3876) (1.3428)
Regression results indicate that SMB is related only to the innovations of aggregate
dividend yield (IDIV), while HML is related only to the innovation of term spread ITERM.
Consistent with Hahn and Lee (2006) and Petkova (2006), the authors provide evidence that
HML is related to a term spread factor. However, we do not find a relationship between
SMB and a default spread factor on CNAS stock market as the authors demonstrated on
U.S. stock market. Instead, we find a significant and negative relationship between SMB
and IDIV. We suspect that SMB shares information with IDIV, while HML shares
information with ITERM. However, the negative relation between HML and ITERM we
found is contrary to Hahn and Lee (2006) and Petkova (2006)’s statement that the HML
factor is positively related to the innovation of the term spread. The regression results show
an insignificant relationship between the two FF factors and innovations of default spread
IDEF, means that IDEF has no explanatory power for SMB or HML.
We next perform the TSRs of contemporaneous innovations of state variables on FF three

factors:
t  ai  bi ( RM ,t  R f )  si SMB  hi HML  i ,t (2.7)
where we denote  t as the innovations of the four state variables and the regression results
are shown in Table 2.5. Consistent with what we observe from the last regressions, IDIV
and SMB, ITERM and HML are negatively and significantly correlated. However, the new
89
Market
evidence shows that both SMB (negatively) and HML (positively) significantly related with
the innovation of one-month T-bill rate IRF.
Table 2.5 Time-series regressions of Contemporaneous innovations of state variables on FF

factors
This table presents time-series regressions’ results of innovations of four state variables on FF three
factors. The adjusted R-squared is represented in the last column and in percentage value. Three FF
factors and four state variables are all demeaned, and the t-statistics presented below the
coefficients are corrected for heteroscedasticity and autocorrelation using the Newey-West
estimator with five lags. The sample period is from December 2006 to May 2015.
Regression: t  ai  bi ( RM ,t  R f )  si SMB  hi HML   i ,t
a b s h Adj. R 2
0.0000 0.0001 -0.0057 -0.0015
IDIV 4.24
(0.0878) (0.0631) (-2.1284) (-0.4699)
-0.0000 0.0014 0.0028 -0.0192
ITERM 7.30
(-0.0135) (0.6146) (0.4466) (-2.9789)
0.0000 -0.0002 0.0013 0.0021
IDEF 2.73
(-0.0122) (-0.1216) (0.3391) (0.4608)
0.0000 -0.0008 -0.0082 0.0108
IRF 5.39
(0.0499) (-0.3148) (-2.2742) (2.4572)
Our findings of HML and ITERM is consistent with Hahn and Lee (2006) who provide
evidence that HML is related to a TERM factor and Petkova (2006) who also find that
ITERM covaries significantly with HML return. Furthermore, they also provide evidence
of a relationship between SMB and DEF factor (or innovations of DEF), which seems not
the case in our study. Inconsistent with Petkova, we find both SMB and HML are
significantly related to IRF, none of FF factors are significantly related to IDEF.
The results from both types of regression suggest that FF factors might be related to the
innovations of state variables, and it is reasonable to examine whether FF factors are
proxies of innovations of state variables on CNAS stock market.
2.3.2 Time-series regressions and results
Following the two-step estimation approach developed by Fama and MacBeth (1973). In
the first step, the TSRs are performed on each of 25 value-weighted Size-B/P portfolios. In
this chapter, the TSRs are performed according to each model (five models) with 102
90
months (December 2006 to May 2015) data of CNAS stock market. We compare five
different factor models, in each model, excess portfolios returns are regressed on different
risk factors. The innovations are obtained from a VAR (1) process.
Ri ,t  R f   i ,t   i ,m  RM ,t  R f     i , j R j ,t  ei ,t
n
(2.8)
1
where, Ri ,t  R f is the excess portfolios return at time t, RM ,t  R f is the excess return on

the market portfolio at the time t, R j ,t are the risk factor (innovations to the state variables
and FF factors) at time t, i , j are the regression coefficients.
2.3.2.1 Five comparative models and time-series regressions
The five TSR models are:
Model 1

(2.9)
 i ,IRF IRF  i ,ISMB ISMB  i ,IHML IHML  ei ,t
Model 2

(2.10)
 i ,IRF IRF  i ,SMB SMB  i ,HML HML  ei ,t
Model 3

(2.11)
 i ,IRF IRF  ei ,t
Model 4
Ri ,t  R f  i  i ,m ( RM ,t  R f )  i ,ISMB ISMB  i ,IHML IHML  ei ,t (2.12)
Model 5
Ri ,t  R f  i  i ,m ( RM ,t  R f )  i ,SMB SMB  i ,HML HML  ei ,t (2.13)
The dependent variables in the five models are the 25 value-weighted Size-B/P portfolios
we constructed in Chapter 1, which remains unchangeable in the regressions. Model 1
91
Market
represents TSRs of the 25 value-weighted Size-B/P portfolios on excess market return,

innovations of state variables and innovations of FF factors SMB and HML (ISMB and
IHML). Similarly, Model 2 represents TSRs on excess market return, innovations of state
variables and the original FF factors SMB and HML. The independent variables in Model 3
are excess market return and four innovations of state variables. In Model 4, the
independent variables are excess market return and innovations of SMB and HML, Model 5
is nothing but the original FF3F Model.
We perform the TSRs of each model during the period December 2006 to May 2015 (102
months), and the regression results are reported separately in Table 2.6, Table 2.7, Table
2.8, Table 2.9 and Table 2.10. The left-half part of each table shows the TSR loadings, the
corresponding t-statistics are reported on the right-half part and are corrected for
autocorrelation and heteroscedasticity using the Newey-West estimator with five lags. At
the end of each table, the adjusted R-square and the residual standard error are also reported.
The numbers in bold indicate the statistical significance at the 5% confidence level.
The regression results of Model 1 are shown in Table 2.6, the market factor still has strong
explanatory power for the average excess stock returns. Both innovations of SMB (ISMB)
and HML (IHML) are important factors in capturing average excess stock returns, all the t-
stats of loadings on ISMB are statistically significant, and almost half of loadings (10 out of
25) on IHML are significant. Specifically, the highest B/P quintile has the most (all five)
significant t-stats, while none of the loadings of the median B/P quintile is significant. The
significant t-stats are concentrated on the lower B/P portfolios, the higher B/P portfolios,
and bigger size portfolios. What’s more, consistent with our previous findings on SMB in
Chapter 1, the slopes on ISMB are systematically related to size, within each B/P quintile,
the loadings decrease as the size increases. The loadings on IHML are also systematically
positively related to B/P ratio, the loadings increase within each size quintile as B/P ratio
increases, except the portfolio which has the smallest size and lowest B/P ratio.
As to the four innovations of state variables IDIV, ITERM, IDEF and IRF, only IDIV has
several slopes that are significant (4 out of 25), none of other innovations have revealed the
relationship with stock returns. Furthermore, none of the loadings on the innovations have
shown a relationship with size or B/P ratio according to our results. Though the adjusted R-
squared are relatively high (with averaged adjusted R-square 0.9018), which means that the
average excess portfolio returns are well explained by Model 1, the majority of statistically
significant of intercepts indicate that the average excess portfolio returns cannot fully
explained by Model 1.
92
Table 2.6 Time-series regression on innovations of state variables and innovations of FF factors (period: December 2006-May 2015)
This table represents the time-series regression results of FF 25 value-weighted Size-B/P portfolios, the independent variables are market
factor, innovations of four state variables and innovations of FF factors. The left-hand part shows the regression coefficients and adjusted R-
square, and the right-hand part are the t-statistics corrected for heteroscedasticity and autocorrelation using the Newey-West estimator with
five lags and the residual standard error. Across rows are the five B/P portfolios and across the columns are the five size portfolios. Numbers
in bold indicate statistical significance at 5% confidence level.
Model 1: Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF  i ,IRF IRF  i ,ISMB ISMB   i ,IHML IHML  ei ,t
Size B/P ratio
L 2 3 4 H L 2 3 4 H
 t  
S 0.0179 0.0252 0.0228 0.0247 0.0227 5.8834 9.3037 9.4438 8.4744 7.8648
2 0.0154 0.0173 0.0159 0.0158 0.0154 4.6410 5.6510 5.7802 5.5111 5.2831
3 0.0115 0.0118 0.0120 0.0135 0.0128 3.2843 4.5352 4.3679 4.4976 4.2713
4 0.0098 0.0083 0.0087 0.0081 0.0065 3.0528 2.8153 2.9950 2.3367 2.3341
B 0.0074 0.0041 0.0022 0.0035 0.0033 2.1024 1.5132 0.6406 1.1473 1.2751
m t  m 
S 1.0209 0.9876 0.9749 1.0120 1.0492 21.3158 26.4098 29.5131 22.7772 29.0068
2 1.0145 1.0152 1.0159 1.0338 1.0269 25.6538 24.7389 25.3328 22.0610 22.7841
3 1.0397 0.9861 1.0308 1.0607 1.0646 14.3735 20.1119 25.5648 25.2300 24.4476
4 0.9429 0.9847 1.0291 1.0863 1.0731 17.6237 25.6068 20.1783 25.8298 23.6878
B 0.9142 1.0668 1.1353 1.1351 1.0957 15.9416 22.5296 21.9166 29.5371 27.3563
 IDIV t   IDIV 
S 3.2751 -2.3102 -0.8156 0.9441 0.0750 0.9032 -0.7082 -0.2985 0.3109 0.0302
2 -6.1561 -0.0931 -1.5454 -5.3985 -3.6960 -1.5625 -0.0273 -0.5056 -1.4915 -1.0674
3 -0.8919 -9.1442 -0.5226 -4.0396 -5.4912 -0.1786 -2.5064 -0.1589 -1.0313 -1.7678
4 -2.8806 -3.5185 -3.2334 -4.3813 -8.5200 -0.7883 -1.1344 -0.8728 -1.5061 -3.0702
B -2.9095 -2.0242 -7.6830 -3.8872 -10.3291 -0.7797 -0.5854 -2.6113 -1.2754 -3.3985
93
94
Table 2.6 Continued
Market
Size B/P ratio

L 2 3 4 H L 2 3 4 H
 ITERM t   ITERM 
S -1.9340 -0.9496 -1.1513 -1.5466 -1.0292 -1.1306 -0.6686 -0.7425 -1.0023 -0.6187
2 -0.6318 -0.7727 -0.2835 -0.8363 -0.1556 -0.4437 -0.4202 -0.2186 -0.5192 -0.1068
3 -1.2174 0.4641 -1.2524 -0.7956 -1.3643 -0.5766 0.2940 -0.8213 -0.5121 -0.9311
4 -1.8504 -0.8149 -0.8775 -0.6636 -0.5776 -0.9658 -0.4176 -0.5017 -0.4357 -0.4730
B -1.7075 -0.0352 -0.2197 0.4397 -0.8593 -0.8929 -0.0202 -0.1133 0.2686 -0.5058
 IDEF t   IDEF 
S 0.8823 -0.9104 0.7130 0.5527 -1.4699 0.2669 -0.3571 0.3543 0.1896 -0.4381
2 -0.9837 -1.9387 -0.4241 1.0599 0.0827 -0.3012 -0.7744 -0.1785 0.4517 0.0353
3 1.4169 0.0828 -0.6519 -0.3042 -0.6642 0.4998 0.0310 -0.2938 -0.1304 -0.2579
4 -1.9631 -1.9086 -0.9741 -2.0648 1.7461 -0.6765 -0.6648 -0.4169 -0.6518 0.6922
B 1.8629 -1.1675 -0.4935 -1.4217 0.6514 0.6610 -0.5129 -0.1856 -0.6323 0.2943
 IRF t   IRF 
S 0.0765 2.0675 -0.0315 1.6425 2.3907 0.0418 1.9204 -0.0285 1.1040 1.6245
2 -0.9933 2.0986 0.3137 0.2620 1.5061 -0.6139 1.3752 0.2821 0.1877 1.1568
3 0.4756 1.4246 -0.5305 0.9008 0.1893 0.2843 1.0858 -0.4079 0.6693 0.1185
4 -0.3925 0.0383 0.7621 2.0425 1.0622 -0.2931 0.0310 0.6112 1.0739 0.6892
B -0.1263 1.7948 2.3551 1.2268 -1.2492 -0.1023 1.1507 1.5545 0.8016 -0.9919
 ISMB t   ISMB 
S 1.7625 1.6844 1.6217 1.7152 1.6767 12.7569 15.5231 16.0337 12.3844 13.5836
2 1.3928 1.4495 1.5397 1.4481 1.5280 10.2365 14.5667 13.1207 11.7784 11.6911
3 1.3956 1.3037 1.3176 1.3120 1.2629 8.5763 11.3764 11.7071 10.7589 10.7399
4 0.9881 1.0498 1.0369 1.1262 0.9986 7.4825 9.2700 8.2822 10.2948 7.3332
B 0.3747 0.4433 0.3966 0.4695 0.3067 2.8519 3.3169 2.5542 3.2099 2.3114
Table 2.6 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
 IHML t   IHML 
S 0.1362 -0.0439 0.0866 0.2263 0.3208 1.0111 -0.4074 0.9580 1.4900 2.7936
2 -0.1188 -0.0955 -0.0416 0.0654 0.5360 -0.7870 -0.8085 -0.3808 0.5474 3.9935
3 -0.3493 -0.2319 0.0153 0.2294 0.4681 -1.7781 -1.5900 0.1339 1.7504 4.1131
4 -0.5403 -0.3710 -0.1307 0.1838 0.5736 -4.3947 -3.3734 -1.0429 1.5094 4.4155
B -0.8560 -0.4135 -0.0589 0.3556 0.9105 -6.2358 -2.8479 -0.3473 2.2694 6.9503
S 0.8845 0.9306 0.9273 0.9088 0.9142 0.0424 0.0313 0.0313 0.0365 0.0359
2 0.9021 0.9090 0.9277 0.9100 0.9121 0.0370 0.0354 0.0318 0.0356 0.0349
3 0.8701 0.9020 0.9035 0.9053 0.9164 0.0444 0.0359 0.0360 0.0362 0.0336
4 0.8690 0.8967 0.8871 0.8954 0.9045 0.0393 0.0355 0.0378 0.0380 0.0353
B 0.8619 0.8920 0.8828 0.9057 0.9088 0.0374 0.0361 0.0395 0.0351 0.0341
95
96
Table 2.7 Time-series regression on innovations of state variables and FF factors, period: December 2006-May 2015
Market
factor, innovations of four state variables and FF factors (SMB and HML). The left-hand part shows the regression coefficients and adjusted
R-square, and the right-hand part are the t-statistics corrected for heteroscedasticity and autocorrelation using the Newey-West estimator with
five lags and the residual standard error. Across rows are the five B/P portfolios and across the columns are the five size portfolios. Numbers
in bold indicate statistical significance at 5% confidence level.
Model 2: Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF  i ,IRF IRF  i ,SMB SMB   i ,HML HML  ei ,t
Size B/P ratio
L 2 3 4 H L 2 3 4 H
 t  
S 0.0157 0.0231 0.0207 0.0225 0.0206 5.1651 7.8780 8.2215 7.4450 6.7707
2 0.0137 0.0155 0.0140 0.0141 0.0134 4.2324 5.2148 5.1658 4.9781 4.5902
3 0.0098 0.0102 0.0104 0.0119 0.0112 2.9827 3.8426 3.8791 4.0150 3.8731
4 0.0086 0.0070 0.0075 0.0067 0.0052 2.7561 2.4128 2.6434 1.9065 1.8877
B 0.0069 0.0036 0.0017 0.0029 0.0029 2.0604 1.2857 0.4953 0.9615 1.1910
m t  m 
S 1.0070 0.9867 0.9648 0.9906 1.0222 18.8441 23.6717 26.7695 18.3013 24.5062
2 1.0189 1.0185 1.0145 1.0256 0.9847 21.8276 21.8706 22.0584 18.9804 18.8420
3 1.0600 0.9997 1.0268 1.0404 1.0267 12.6328 17.1738 22.6575 21.8938 20.6672
4 0.9797 1.0078 1.0361 1.0696 1.0287 17.0552 22.5750 18.4362 22.5940 19.4547
B 0.9745 1.0946 1.1381 1.1080 1.0286 16.1810 20.3107 18.8399 25.4167 21.6820
S 3.3139 -2.4932 -0.9153 0.8616 0.0114 0.8356 -0.7328 -0.3381 0.2670 0.0039
2 -6.1348 -0.2314 -1.6289 -5.4936 -3.7849 -1.5570 -0.0650 -0.5194 -1.4663 -1.0420
3 -0.8089 -9.4098 -0.6392 -4.1060 -5.4866 -0.1633 -2.4484 -0.2017 -1.0900 -1.7971
4 -3.0447 -3.6051 -3.3329 -4.4353 -8.5388 -0.8477 -1.1314 -0.9357 -1.5256 -3.3468
B -2.9486 -2.1363 -7.8680 -3.9625 -10.3656 -0.7906 -0.6095 -2.6095 -1.2874 -3.5048
Table 2.7 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
S -1.9810 -0.9542 -1.1701 -1.5072 -1.0541 -1.2324 -0.6413 -0.7889 -1.0040 -0.6426
2 -0.6175 -0.7760 -0.2606 -0.8462 -0.1644 -0.4141 -0.4033 -0.1948 -0.5104 -0.1106
3 -1.1817 0.4682 -1.2758 -0.7700 -1.3181 -0.5517 0.2794 -0.8090 -0.4772 -0.9178
4 -1.8697 -0.7666 -0.8908 -0.6319 -0.5493 -0.9516 -0.3839 -0.4903 -0.4003 -0.4291
B -1.7003 0.0132 -0.1771 0.4690 -0.7978 -0.8758 0.0075 -0.0902 0.2802 -0.4679
S 0.8691 -0.9023 0.7136 0.5664 -1.4726 0.2717 -0.4030 0.3761 0.2102 -0.4831
2 -0.9813 -1.9325 -0.4143 1.0624 0.0850 -0.3198 -0.8341 -0.1896 0.4802 0.0384
3 1.4213 0.0972 -0.6516 -0.2947 -0.6534 0.5064 0.0371 -0.3041 -0.1316 -0.2669
4 -1.9594 -1.8927 -0.9723 -2.0545 1.7539 -0.7057 -0.6949 -0.4183 -0.6737 0.7012
B 1.8666 -1.1502 -0.4739 -1.4109 0.6680 0.6824 -0.5004 -0.1778 -0.6325 0.3002
S 0.1036 2.0274 -0.0470 1.6084 2.3858 0.0538 1.8046 -0.0411 1.1023 1.6519
2 -0.9939 2.0682 0.2857 0.2441 1.4892 -0.6161 1.4378 0.2604 0.1820 1.1181
3 0.4807 1.3623 -0.5481 0.8757 0.1725 0.2921 0.9731 -0.4220 0.6623 0.1114
4 -0.4226 -0.0002 0.7445 2.0178 1.0469 -0.3170 -0.0001 0.6285 1.0351 0.7011
B -0.1380 1.7504 2.2963 1.1982 -1.2813 -0.1105 1.0989 1.4897 0.7897 -1.0832
 SMB t   SMB 
S 1.7645 1.6685 1.6126 1.7092 1.6705 13.2778 16.2937 16.1977 12.9987 13.9200
2 1.3951 1.4375 1.5331 1.4396 1.5201 10.7893 14.6810 13.8857 11.9651 12.5936
3 1.4038 1.2810 1.3069 1.3070 1.2646 8.8594 11.0313 12.3096 11.6093 12.0041
4 0.9735 1.0438 1.0279 1.1225 0.9978 7.5556 9.4158 8.4995 11.3355 7.9670
B 0.3715 0.4351 0.3819 0.4638 0.3054 2.8146 3.2339 2.4492 3.3447 2.4351
97
98
Market
Table 2.7 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
 HML t   HML 
S 0.1257 -0.0511 0.0786 0.2334 0.3125 0.8636 -0.4580 0.8007 1.4646 2.5827
2 -0.1145 -0.1009 -0.0386 0.0598 0.5309 -0.7393 -0.7904 -0.3439 0.4734 3.9391
3 -0.3375 -0.2397 0.0056 0.2336 0.4798 -1.6685 -1.5336 0.0470 1.7308 4.1866
4 -0.5506 -0.3618 -0.1373 0.1899 0.5801 -4.2622 -3.0208 -1.0487 1.5061 4.5625
B -0.8555 -0.4051 -0.0544 0.3605 0.9247 -5.9363 -2.6389 -0.3089 2.2640 7.2142
S 0.8900 0.9295 0.9285 0.9105 0.9165 0.0414 0.0316 0.0310 0.0362 0.0354
2 0.9051 0.9087 0.9286 0.9108 0.9137 0.0364 0.0354 0.0317 0.0355 0.0346
3 0.8738 0.8984 0.9037 0.9063 0.9197 0.0438 0.0366 0.0360 0.0360 0.0329
4 0.8687 0.8956 0.8871 0.8962 0.9067 0.0394 0.0357 0.0378 0.0378 0.0349
B 0.8632 0.8906 0.8816 0.9059 0.9126 0.0372 0.0363 0.0397 0.0350 0.0334
In Table 2.7, which reports the regression results of Model 2, we replace ISMB and IHML
by the original FF factors SMB and HML, the other risk factors remain the same. The
results are much like those of Model 1. It is not difficult to understand the similar results
since SMB and HML are highly correlated to their innovations (refer to Table 2.3).
Table 2.8 displays the TSR results on innovations of four selected state variables in
addition to excess market return (Model 3) over our sample period. Without FF factors or
their innovations in the presence of regressions, IDIV plays an important role in capturing
time-series variation of average excess returns with 23 out of 25 loadings are statistically
significant at 5% confidence level (t-stats are corrected for heteroscedasticity and
autocorrelation using the Newey-West estimator with five lags). Furthermore, the loadings
on IDIV are all negative, within each B/P quintile, most of the loadings increase from
smaller size portfolios to bigger size portfolios except the smallest size quintile, and the
biggest size portfolios tend to have higher returns.
None of the loadings on ITERM or on IDEF is significant, only one out of 25 loadings on
IRF is significant. However, consistent with Petkova (2006) that the loadings on ITERM
are related to B/P ratio, within each size quintile, the higher B/P ratio portfolios tend to
have smaller loadings, at least for the three higher B/P portfolios. Both Hahn and Lee (2006)
and Petkova (2006) find the innovation of default spread is systematically related to size,
though, we find no such a relationship between loadings on IDEF and firm size in this
study. IRF seems to fluctuate with size, bigger size portfolios tend to have smaller loadings
on IRF within each B/P quintile.
Comparing the adjusted R-squares of Model 1, Model 2 and Model 3, We notice that the
adjusted R-square of Model 3 (with averaged adjusted R-square 0.7228) are much lower
than those of Model 1 (with averaged adjusted R-square 0.9018) and Model 2 (with
averaged adjusted R-square 0.9021), which indicate the lower descriptive power of Model 3,
which contains only innovations of four state variables without FF factors (or innovations
of FF factors).
Table 2.9 reports the TSR results on excess market return and innovations of SMB and
HML (ISMB and IHML). Loadings on ISMB are all statistically significant at 5%
confidence level and have an inverse relationship with size across each B/P quintile.
Almost half of loadings on IHML (10 out of 25) are significant, and consistent with results
of Model 1, the loadings are positively related to B/P ratio apart from the portfolio with the
smallest size and lowest B/P ratio. The adjusted R-squares are all around 90% (with
averaged adjusted R-square 0.9030).
99
Table 2.8 Time-series regression on innovations of state variables, period: December 2006-May 2015
100
Market
factor, innovations of four state variables. The left-hand part shows the regression coefficients and adjusted R-square, and the right-hand part
are the t-statistics corrected for heteroscedasticity and autocorrelation using the Newey-West estimator with five lags and the residual standard
error. Across rows are the five B/P portfolios and across the columns are the five size portfolios. Numbers in bold indicate statistical
significance at 5% confidence level.
Model 3: Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IDIV IDIV  i ,ITERM ITERM  i ,IDEF IDEF  i ,IRF IRF  ei ,t
Size B/P ratio
L 2 3 4 H L 2 3 4 H
 t  
S 0.0179 0.0252 0.0228 0.0247 0.0227 3.3759 4.3212 4.2855 4.1337 4.0746
2 0.0154 0.0173 0.0159 0.0158 0.0153 2.8798 3.1962 2.9853 3.0231 2.8955
3 0.0115 0.0118 0.0120 0.0135 0.0128 2.1280 2.4358 2.5277 2.7405 2.7502
4 0.0098 0.0083 0.0087 0.0081 0.0065 1.9630 1.7633 1.9847 1.6390 1.5922
B 0.0074 0.0041 0.0022 0.0035 0.0033 1.6861 1.1976 0.6040 1.0010 0.8332
m t  m 
S 1.0209 0.9876 0.9749 1.0120 1.0492 9.8737 10.7385 11.9462 10.7362 10.9240
2 1.0145 1.0152 1.0159 1.0338 1.0269 13.1740 12.4564 12.2129 12.2937 11.9734
3 1.0397 0.9861 1.0308 1.0607 1.0646 9.3296 11.7160 14.1004 14.4893 14.0225
4 0.9429 0.9847 1.0291 1.0863 1.0731 11.0381 13.6882 13.6469 15.2915 17.2509
B 0.9142 1.0668 1.1353 1.1351 1.0957 10.1041 15.7473 19.4823 24.9320 23.6765
S -17.9844 -22.8712 -20.4309 -19.6134 -19.8813 -2.0760 -2.6197 -2.7405 -2.5233 -2.4337
2 -23.2737 -17.8676 -20.3418 -22.9303 -21.5411 -3.0490 -2.2935 -2.6387 -2.8917 -2.9174
3 -18.3666 -25.3354 -16.5371 -19.6851 -20.2047 -2.4629 -3.3987 -2.6421 -3.4911 -3.4613
4 -15.6639 -16.8155 -16.0357 -17.8299 -19.8692 -2.3220 -2.6005 -3.1312 -3.4911 -4.7184
B -8.6693 -7.9993 -12.5921 -9.1024 -12.7861 -1.5932 -1.9900 -3.4977 -2.4445 -2.8244
Table 2.8 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
S 0.3304 1.9408 1.0936 0.2657 0.3266 0.0724 0.4942 0.2776 0.0649 0.0772
2 2.1025 1.9554 2.3645 1.2180 0.0632 0.6040 0.6455 0.6404 0.3923 0.0203
3 2.4832 3.5273 0.8015 0.3557 -1.2881 0.6216 1.0322 0.2751 0.1340 -0.4928
4 1.9927 2.4208 1.3345 0.3795 -1.3664 0.6967 0.8356 0.6054 0.1596 -0.7351
B 2.4672 2.4032 0.6637 -0.2899 -4.1661 1.0137 1.4308 0.4201 -0.2156 -1.7841
S 2.5025 0.8242 2.2451 2.0290 -0.1335 0.4632 0.1414 0.4629 0.4119 -0.0252
2 0.5390 -0.3841 1.1631 2.4408 1.0397 0.1119 -0.0807 0.2429 0.5707 0.2581
3 3.1892 1.6373 0.6519 0.7647 0.0999 0.6292 0.3601 0.1576 0.1926 0.0268
4 -0.3945 -0.4595 0.2047 -1.1332 2.1325 -0.1019 -0.1073 0.0605 -0.2711 0.6828
B 3.1549 -0.2805 -0.0330 -1.3321 -0.0161 0.9075 -0.1004 -0.0128 -0.5574 -0.0055
S -5.6662 -3.7575 -5.3902 -3.7645 -2.7022 -1.8233 -1.0569 -1.7737 -1.1558 -0.8596
2 -5.9695 -3.0257 -5.0135 -4.5460 -2.6636 -1.8999 -0.8930 -1.7084 -1.5258 -1.0902
3 -4.9562 -3.4668 -4.9910 -3.1261 -3.2084 -1.5812 -1.1563 -1.8576 -1.2471 -1.4731
4 -4.8055 -4.2573 -3.0243 -1.4397 -1.2308 -2.1350 -1.6505 -1.3672 -0.5494 -0.6474
B -3.0597 -0.5163 0.8896 0.3151 -0.5325 -1.4298 -0.2571 0.5533 0.1940 -0.2992
S 0.6034 0.6311 0.6478 0.6313 0.6604 0.0786 0.0722 0.0689 0.0734 0.0713
2 0.6877 0.6760 0.6755 0.6973 0.6906 0.0660 0.0667 0.0675 0.0654 0.0655
3 0.6480 0.6901 0.7159 0.7369 0.7618 0.0731 0.0639 0.0618 0.0603 0.0567
4 0.6856 0.7266 0.7538 0.7710 0.8019 0.0609 0.0577 0.0559 0.0562 0.0508
B 0.7397 0.8422 0.8661 0.8825 0.8471 0.0514 0.0436 0.0422 0.0391 0.0442
101
102
Table 2.9 Time-series regression on innovations of FF factors, period: December 2006-May 2015
Market
factor and innovations of FF factors. The left-hand part shows the regression coefficients and adjusted R-square, and the right-hand part is the
t-statistics corrected for heteroscedasticity and autocorrelation using the Newey-West estimator with five lags and the residual standard error.
Across rows are the five B/P portfolios and across the columns are the five size portfolios. Numbers in bold indicate statistical significance at
5% confidence level.
Model 4: Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,ISMB ISMB  i ,IHML IHML  ei ,t
Size B/P ratio
L 2 3 4 H L 2 3 4 H
 t  
S 0.0179 0.0252 0.0228 0.0247 0.0227 5.8570 9.1204 9.6651 8.7478 7.9806
2 0.0154 0.0173 0.0159 0.0158 0.0154 4.6173 5.7955 5.7843 5.5248 5.2366
3 0.0115 0.0118 0.0120 0.0135 0.0128 3.3790 4.2925 4.4080 4.5435 4.1679
4 0.0098 0.0083 0.0087 0.0081 0.0065 3.0289 2.7980 3.0563 2.2689 2.1847
B 0.0074 0.0041 0.0022 0.0035 0.0033 2.1748 1.4854 0.6195 1.1330 1.2000
m t  m 
S 1.0209 0.9876 0.9749 1.0120 1.0492 20.4422 23.2634 28.1865 21.1957 26.2015
2 1.0145 1.0152 1.0159 1.0338 1.0269 22.4253 22.7165 24.1228 18.9534 20.1681
3 1.0397 0.9861 1.0308 1.0607 1.0646 13.6161 16.1596 25.1005 22.5499 21.2923
4 0.9429 0.9847 1.0291 1.0863 1.0731 16.6857 23.0171 18.7585 22.5114 18.6476
B 0.9142 1.0668 1.1353 1.1351 1.0957 14.7199 20.9350 18.0653 26.6419 20.0845
 ISMB t   ISMB 
S 1.7388 1.6770 1.6245 1.6927 1.6518 13.0825 14.7265 16.3813 12.2473 12.8010
2 1.4338 1.4281 1.5447 1.4764 1.5367 10.1991 13.8343 13.1178 10.9691 11.7009
3 1.3954 1.3464 1.3208 1.3254 1.2887 9.3564 10.0790 12.2018 11.2270 11.1690
4 1.0006 1.0653 1.0457 1.1303 1.0400 7.4170 8.6293 8.5051 10.0121 7.4292
B 0.3904 0.4387 0.4205 0.4813 0.3756 2.9629 3.4108 2.5686 3.3733 2.4881
Table 2.9 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
 IHML t   IHML 
S 0.1717 -0.0009 0.1117 0.2744 0.3640 1.2480 -0.0076 1.1739 1.7731 3.0342
2 -0.1098 -0.0613 -0.0310 0.0956 0.5617 -0.6870 -0.4863 -0.2801 0.7446 4.0172
3 -0.3156 -0.2106 0.0334 0.2606 0.5041 -1.5762 -1.4259 0.2884 1.9343 4.0999
4 -0.5081 -0.3532 -0.1021 0.2217 0.6140 -4.0710 -3.2319 -0.8287 1.7259 4.4300
B -0.8153 -0.3924 -0.0176 0.3636 0.9314 -5.8388 -2.7476 -0.1015 2.3487 6.5998
S 0.8875 0.9313 0.9296 0.9103 0.9153 0.0419 0.0312 0.0308 0.0362 0.0356
2 0.9041 0.9107 0.9305 0.9117 0.9144 0.0366 0.0350 0.0312 0.0353 0.0344
3 0.8743 0.9016 0.9070 0.9078 0.9177 0.0437 0.0360 0.0354 0.0357 0.0333
4 0.8727 0.8998 0.8906 0.8970 0.9033 0.0388 0.0350 0.0372 0.0377 0.0355
B 0.8645 0.8951 0.8832 0.9083 0.9064 0.0371 0.0356 0.0394 0.0346 0.0346
103
Market
The TSR results of Model 5 (the original FF3F Model) in Table 2.10 are extremely like
those of Model 4 (Table 2.9), FF3F Model still explains the time-series average excess
returns well on CNAS stock market during December 2006 to May 2015. All the t-stats of
loadings on excess market return and SMB are highly significant, and 10 out of 25 loadings
on HML are significant as well, the adjusted R-squared are all around 90%, similar as those
in Table 2.9. However, the majority of significantly intercept continues indicate that the
average excess stock returns on CNAS stock market cannot fully capture by FF3F Model.
2.3.2.2 Brief comparison and summary
Comparing Model 1 and Model 2, the difference is that we include innovations of FF

factors (ISMB and IHML) or FF factors themselves (SMB and HML) in addition to excess
market return and innovations of state variables. The results tell no remarkable differences
between the TSRs of the two models. It is revealed that FF factors and their innovations
have approximately the same explanatory power for stock returns on CNAS stock market.
Comparing Model 3 with Model 2 (or Model 1), the regression results suggest that the
regression model contains only the innovations of state variables without FF factors (or
their innovations) as independent variables dramatically reduce the explanatory power of
the model. The connections between innovations and size or B/P ratio (IDIV and size,
ITERM and B/P ratio, and IRF and size) disappear when the regressions are performed
with FF factors (or their innovations). Furthermore, in the presence of FF factors (or their
innovations), the innovations of state variables - especially IDIV - seem not able to explain
time-series variation of expected stock returns.
Comparing Model 4 with Model 1 and Model 5 with Model 2, including innovations of
state variables or not in the regression model do not change the performance of FF factors
(innovations of FF factors) in capturing time-series variation of expected stock returns. The
innovations of state variables that describe time variation in investment opportunities have
little or marginal effect when adding them into FF3F Model or model with innovations of
FF factors as independent variables. Comparing Model 4 and Model 5, the results prove
again the high correlation between FF factors and their innovations, since the regressions
results are much alike from both models.
We can conclude that FF factors (or innovations of FF factors) explain the time-series
variation of expected stock returns well, with or without the innovations of state variables
are included in the model. When the four innovations of state variables are regressed alone,
104
Table 2.10 Time-series regression on FF3F model, period: December 2006-May 2015
This table shows the time-series regression results of FF 25 value-weighted Size-B/P portfolios on FF3F model. The left-hand part shows the
regression coefficients and adjusted R-square, and the right-hand part are the t-statistics corrected for heteroscedasticity and autocorrelation
using the Newey-West estimator with five lags and the residual standard error. Across rows are the five B/P portfolios and across the columns
are the five size portfolios. Numbers in bold indicate statistical significance at 5% confidence level.
Model 5: Ri ,t  R f   i   i ,m ( RM ,t  R f )  i ,SMB SMB  i ,HML HML  ei ,t
Size B/P ratio
L 2 3 4 H L 2 3 4 H
 t  
S 0.0158 0.0232 0.0207 0.0226 0.0207 5.1197 7.5733 8.4139 7.5664 6.7025
2 0.0136 0.0156 0.0140 0.0140 0.0134 4.2131 5.2460 5.1461 4.9841 4.5060
3 0.0098 0.0101 0.0104 0.0119 0.0112 3.0746 3.6019 3.9167 4.0346 3.7754
4 0.0086 0.0070 0.0075 0.0067 0.0052 2.7130 2.3837 2.6843 1.8375 1.7431
B 0.0069 0.0036 0.0017 0.0029 0.0028 2.1252 1.2625 0.4714 0.9402 1.1059
m t  m 
S 1.0045 0.9837 0.9630 0.9873 1.0192 18.6519 20.8981 25.2693 17.5126 23.0016
2 1.0182 1.0161 1.0138 1.0234 0.9828 18.9259 20.4411 21.1811 16.3995 16.7307
3 1.0577 0.9981 1.0255 1.0382 1.0242 12.0269 13.9348 22.5043 19.2727 17.8041
4 0.9773 1.0066 1.0341 1.0670 1.0259 16.2656 20.6871 17.1025 19.5467 15.5571
B 0.9716 1.0932 1.1352 1.1075 1.0271 14.8421 18.9966 15.4809 22.7688 16.5446
 SMB t   SMB 
S 1.7402 1.6623 1.6158 1.6876 1.6460 13.4101 14.8754 16.2425 12.6110 12.8491
2 1.4354 1.4172 1.5389 1.4680 1.5291 10.4139 13.5197 13.4322 10.8677 12.2027
3 1.4029 1.3250 1.3107 1.3207 1.2901 9.5778 9.6442 12.4894 11.7651 12.0529
4 0.9867 1.0599 1.0372 1.1269 1.0387 7.3400 8.5896 8.5953 10.6518 7.8389
B 0.3871 0.4315 0.4069 0.4762 0.3738 2.9199 3.3151 2.4676 3.4661 2.5677
105
106
Market
Table 2.10 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
 HML t   HML 
S 0.1616 -0.0090 0.1034 0.2797 0.3554 1.0902 -0.0784 1.0136 1.7539 2.8807
2 -0.1062 -0.0674 -0.0289 0.0893 0.5561 -0.6579 -0.5031 -0.2588 0.6766 4.0611
3 -0.3054 -0.2195 0.0237 0.2635 0.5139 -1.5122 -1.4237 0.2018 1.9478 4.2881
4 -0.5189 -0.3457 -0.1092 0.2262 0.6186 -4.0547 -2.9736 -0.8671 1.7360 4.6804
B -0.8159 -0.3856 -0.0152 0.3674 0.9433 -5.6547 -2.5790 -0.0865 2.3783 6.9345
S 0.8927 0.9303 0.9308 0.9119 0.9175 0.0409 0.0314 0.0305 0.0359 0.0352
2 0.9070 0.9104 0.9313 0.9123 0.9159 0.0360 0.0351 0.0310 0.0352 0.0341
3 0.8778 0.898 0.9071 0.9088 0.9209 0.0431 0.0367 0.0353 0.0355 0.0327
4 0.8723 0.8988 0.8905 0.8978 0.9055 0.0388 0.0351 0.0373 0.0375 0.0351
B 0.8657 0.8938 0.8819 0.9086 0.9100 0.0369 0.0358 0.0397 0.9086 0.0339
only IDIV has explanatory power for average return; while in the presence of FF factors (or
innovations of FF factors), IDIV loses its capability in capturing time-series variation of
expected stock returns. FF factors contain the information of innovations of aggregate
dividend yield (IDIV) on CNAS stock market over our research period. In the next section,
we will perform the CSRs among the comparing five models, and examine whether
innovations of state variables have explanatory power in cross-section variation of excess
stock returns, furthermore, whether FF factors proxy for the innovations of state variables
in the cross-section.
2.4 Cross-sectional validation of five comparing models
In this section, following Petkova (2006), we examine cross-section validation that FF

factors proxy for innovations of state variables on CNAS stock market over our sample
period. The objective is to test whether an asset’s loadings with respect to these risk factors
are important determinants of its average return.
In the second pass of Fama and MacBeth’s approach, the CSRs are performed by
regressing the portfolios’ excess returns on the estimated betas obtained from the first step
of Fama-MacBeth’s two-stage approach, for a given date.
n
Ri ,t  R f   0   m ˆ i ,m    j ˆ i , j   i ,t (2.14)
1
where, the independent variables, ̂ s are the loadings which have been estimated from the
TSRs and stand for exposures to the corresponding risk factor.  s are the coefficients of the
CSRs and stand for the reward for bearing the risk of that factor (‘price of risk’ or ‘risk
premium’). “If loadings with respect to innovations in a state variable are important
determinants of average returns, then there should be a significant price of risk associated
with that state variable” 27.
As documented in Chapter 1, the Fama and MacBeth (1973) two-stage approach has the
classical Errors-in-variables (EIV) problem, thus, we also report Shanken (1992)’s adjusted
t-stats (SH t-stats) in the CSRs.
27
Petkova (2006)
107
Market
Considering our research, for each month, we regress the excess returns of 25 Size-B/P
portfolios on the estimated betas that are obtained from the TSRs in the previous section.
Since we have 102 months over our sample period (December 2006 to May 2015), 102
CSRs have been performed using OLS regressions, then the regression constants and the
coefficients have been averaged over the 102 estimations.
We perform the cross-section regressions of five models corresponding to the five time-
series models we have examined in the previous section:
Model 1#
Ri ,t  R f   0   m ˆ i ,m   IDIV ˆ i ,IDIV   ITERM ˆ i ,ITERM   IDEF ˆ i ,IDEF   IRF ˆ i ,IRF

(2.15)
 ISMB ˆ i ,ISMB   IHML ˆ i ,IHML   i ,t
Model 2#
Ri ,t  R f   0   m ˆ i ,m   IDIV ˆ i ,IDIV   ITERM ˆ i ,ITERM   IDEF ˆ i ,IDEF   IRF ˆ i ,IRF

(2.16)
 ˆ SMB i ,SMB  ˆHML i ,HML  i ,t
Model 3#
Ri ,t  R f   0   m ˆ i ,m   IDIV ˆ i ,IDIV   ITERM ˆ i ,ITERM   IDEF ˆ i ,IDEF   IRF ˆ i ,IRF   i ,t (2.17)
Model 4#
Ri ,t  R f   0   m ˆ i ,m   ISMB ˆ i ,ISMB   IHML ˆ i ,IHML   i ,t (2.18)
Model 5#
Ri ,t  R f   0   m ˆ i ,m   SMB ˆ i ,SMB   HML ˆ i ,HML   i ,t (2.19)
Ri ,t  R f are the same 25 Size-B/P portfolios as in the TSRs, and for each model, the
estimated betas are obtained from the TSRs of the corresponding model (for example the
estimated betas in Model 1# are the TSR coefficients of Model 1). Model 1# is the CSR on
the loadings of excess market return, four innovations of state variables and innovations of
FF factors. Model 2# represents the CSR on the loadings of excess market return, four
innovations of state variables and FF factors. Model 3# represents the CSR on the loadings
of excess market return and four innovations of state variables. Model 4# is the CSR of
market factor and innovations of SMB and HML, while Model 5# is nothing but the CSR
of FF3F Model.
108
The CSRs results of the five models are reported in Table 2.11. Across the rows we report
the regressions coefficients and the corresponding t-statistics below each coefficient. The
adjusted R-squares are reported in the last column. The numbers in bold indicate statistical
significance at the 5% confidence level. We analyze the results by comparing the models:
- Comparing Model 1# and Model 2#, the loadings on exposure to market factor ̂ m
(with FM t-stats -2.9752 in Model 1# and -3.0838 in Model 2#) and exposure to the
innovation of one-month T-bill rate ̂ IRF (with FM t-stats 2.2365 in Model 1# and
2.3197 in Model 2#) are statistically significant in both models no matter regress
with exposure to FF factor or their innovations. However, under the EIV correction,
the significance of t-stats only exists for market factor. ISMB in Model 1# and SMB
in Model 2# are both significantly priced, robust to the EIV adjustment; while the
exposures to IHML or HML are not significant variables in the cross section. None
of the exposures to innovations of state variables is significantly priced in
explaining the cross-sectional excess portfolios’ returns adjusted to EIV problem.
The results show that whether we regress innovations of state variables with SMB
and HML (or with ISMB and IHML) does not change the empirical outcomes.
The very similar results obtained from Model 4# and Model 5# prove again that the
FF factors (SMB and HML) and their innovations (ISMB and IHML) make no big
difference between the CSR results. The market factor and ISMB are priced in
Model 4#, correspondingly, the market factor and SMB are significantly priced in
Model 5#. It is noteworthy that Model 5# is exactly the CSR of the original FF3F
Model, the market beta is an important determinant in explaining the cross-sectional
variation of the excess portfolio returns during the period December 2006 to May
2015, and the estimated coefficient of the loadings on market beta is negative (-
0.0334) with FM t-stats -2.1285 and Shanken adjusted t-stats is -3.3191. Consistent
with previous findings, size factor SMB is positively significantly priced and value
factor HML still remains not significantly priced in the cross section of portfolio
returns. This finding indicates that the ability of market factor and SMB in
capturing cross-sectional variation of stock returns is robust at least over our
research periods28.
28
We examine the cross-sectional validation of FF3F Model on Chinese A-share stock market
during July 2004 to May 2015, and we find that there exists market premium and size premium over
this sample period.
109
Table 2.11 Cross-sectional regressions of five comparative models
110
Market
This table reports the Fama-MacBeth cross-sectional regressions using excess returns of 25 Size-B/P portfolios on Chinese A-share stock
market, the independent variables are the estimated loadings of time-series regressions from the first stage of Fama-MacBeth approach. We
perform the cross-sectional regressions for the following five models:
Model 1#: Ri ,t  R f   0   m ˆ i ,m   IDIV ˆ i ,IDIV   ITERM ˆ i ,ITERM   IDEF ˆ i ,IDEF   IRF ˆ i ,IRF   ISMB ˆ i ,ISMB   IHML ˆ i ,IHML   i ,t
Model 2#: Ri ,t  R f   0   m ˆ i ,m   IDIV ˆ i ,IDIV   ITERM ˆ i ,ITERM   IDEF ˆ i ,IDEF   IRF ˆ i ,IRF   SMB ˆ i ,SMB   HML ˆ i ,HML   i ,t
Model 3#: Ri ,t  R f   0   m ˆ i ,m   IDIV ˆ i ,IDIV   ITERM ˆ i ,ITERM   IDEF ˆ i ,IDEF   IRF ˆ i ,IRF   i ,t
Model 4#: Ri ,t  R f   0   m ˆ i ,m   ISMB ˆ i ,ISMB   IHML ˆ i ,IHML   i ,t
Model 5#: Ri ,t  R f   0   m ˆ i ,m   SMB ˆ i ,SMB   HML ˆ i ,HML   i ,t
For each model, the cross-sectional regressions are performed on each month (102 months) during the period December 2006 to May 2015,
then the 102 constants and coefficients have been averaged. The first row within each model reports the regression coefficients (gammas), and
the corresponding Fama-MacBeth (FM) t-stats and Shanken adjusted (SH) t-stats are reported in the brackets below the coefficients across the
second row and third row, separately. Numbers in bold indicate statistical significance at 5% confidence level. The averaged adjusted R-
squared are reported in the last column in percentage form.
2
Adj. R
Model 0 m  IDIV  ITERM  IDEF  IRF  ISMB  IHML  SMB  HML
(%)
Model 1# 0.0539 -0.0516 0.0003 -0.0001 0.0002 0.0012 0.0090 0.0048
FM t-stats (3.5969) (-2.9752) (1.8283) (-0.1314) (0.6759) (2.2365) (2.1616) (1.2536) 56.22
SH t-stats (-4.9132) (1.2735) (-0.0885) (0.4741) (1.6731) (2.0987) (1.1875)
Model 2# 0.0560 -0.0534 0.0003 -0.0000 0.0002 0.0012 0.0088 0.0010

FM t-stats (3.7506) (-3.0838) (1.8186) (-0.0297) (0.6181) (2.3197) (2.1219) (0.2678) 56.05
SH t-stats (-5.0751) (1.2436) (-0.0196) (0.4279) (1.7166) (2.0600) (0.2481)
Model 3# 0.0196 -0.0229 -0.0007 -0.0004 0.0005 -0.0008

FM t-stats (1.1463) (-1.2504) (-2.9932) (-0.5596) (1.3557) (-1.2350) 46.66
SH t-stats (-2.1532) (-2.0752) (-0.3679) (1.0195) (-1.0283)
2
Adj. R
Model 0 m  IDIV  ITERM  IDEF  IRF  ISMB  IHML  SMB  HML
(%)
Model 4# 0.0313 -0.0330 0.0118 0.0022
FM t-stats (2.3868) (-2.1064) (2.8090) (0.5808) 52.01
SH t-stats (-3.2828) (2.8039) (0.5827)
Model 5# 0.0318 -0.0334 0.0117 -0.0002

FM t-stats (2.4221) (-2.1285) (2.7876) (-0.0654) 51.99
SH t-stats (-3.3191) (2.7877) (-0.0643)
111
Market
- Comparing Model 3# with Model 1# or Model 2# (we take the pair of Model 3# and
Model 2# as example since the results of Model 1# and Model 2# are quite similar),
when the CSRs are performed on the loadings of market factor and innovations of
four state variables (Model 3#) eliminating SMB and HML (or ISMB and IHML),
only loadings on excess market return and on IDIV is significantly priced (EIV
adjusted). Never IDEF or ITERM or IRF has significant loadings in CSRs
regardless of regressing with only innovations of state variables (Model 3#) or with
FF factors (Model 2#). Especially, though the loading on IDIV is the only
significant determinant of average returns, in the presence of FF factors or their
innovations, the significance of loading on IDIV disappears. The information
contained in IDIV seems totally captured by the market factor and size factor in
explaining the cross-sectional variation of average portfolio returns.
- Comparing Model 4# with Model 1# and Model 5# with Model 2#, the analyze is
much similar for both pairs of models, thus we take Model 5# and Model 2# for
example. In Model 2#, none of the loadings on IDIV, ITERM, IDEF or IRF is
important determinants of average returns; and the presence of innovations of the
four state variables do not drive FF factors out.
- Comparing all the five models, inconsistent with Petkova (2006), the loadings on
the excess market return are always negatively statistically significant on CNAS
stock market, which suggest that there exists robust negative market premium. The
exposures to ITERM, IDEF and IRF are not significant variables by all means in
cross section (Petkova finds that loading of ITERM is a significant factor in the
cross section of 25 portfolios). SMB or ISMB is an important determinant in
explaining the variation of cross-sectional stock returns but HML or IHML is not;
the existence of size premium and the lack of value premium are robust to the
different research periods on CNAS stock market. In terms of the average adjusted
R-square of the five models, it is obvious that Model 3# has the lowest one
(46.66%). While the other models explain relatively a larger percentage of the
cross-sectional variation in average returns of portfolios than Mode 3, in which the
independent variables are only market excess return and innovations of state
variables.
We summarize the findings of CSR in Table 2.11 and answering the questions proposed at
the beginning of this section. The presence of the four innovations of state variables that
forecast future investment opportunities does not drive FF factors out. The information
contained in the innovation of aggregate dividend yields IDIV seems totally captured by the
combination of market beta and SMB (or ISMB). Though the model involves both FF
112
factors and innovations of state variables as risk factors performs slightly better than the
original FF3F Model (considering the averaged adjusted R-square terms), FF factors might
have played a limited role in capturing alternative investment opportunities proxied by
innovations of state variables.
A supplement conclusion of Chapter 1: market beta and SMB are able to explain the cross-
sectional variation of average stock returns except for the value factor HML over the period
December 2006 to May 2015 on CNAS stock market. Furthermore, it seems that there exist
robust negative market premium and positive size premium but no value premium on
CNAS stock market, which is independent of research periods.
2.5 Conclusions
This paper investigates the explanatory ability of the innovations of four macroeconomic
variables: aggregate dividend yield, one-month T-bill rate, term spread and default spread
on CNAS stock market, using monthly data during the period December 2006 to May 2015
(102 months). To examine whether the innovations of the selected four predictive variables
are able in capturing average excess stock returns in both time-series and cross-section,
further whether FF factors SMB and HML proxy for the innovations of state variables that
describe future investment opportunities on CNAS stock market, the TSRs and CSRs are
performed on FF 25 Size-B/P portfolios on five comparing models separately.
Results from the TSR indicate that FF factors don’t lose their explanatory power no matter
when examined alone or in combination with innovations of state variables on CNAS stock
market. When regressed alone, the innovations of selected state variables do not have the
ability in capturing average stock returns except IDIV, which might indicate that the FF
factors totally capture the information of IDIV in explaining time-series stock returns.
Consistent with literature, we find innovation of term spread is related to B/P ratio, while
inconsistently, we find no systematical relationship between innovation of default spread
and firm size, instead, we find IDIV is related to size on CNAS stock market.
We conclude from the CSRs, the original FF factors (market beta and SMB) has ability in
capturing the cross-sectional variation of excess stock returns and there are significant
market risk premium and size premium during the period December 2006 to May 2015.
Inconsistent with studies on U.S. stock market, in the presence of four innovations of state
variables, FF factors do not lose their ability in capturing cross-sectional variations of
excess stock returns on Chinese stock market. We find the information contained in the
113
Market
innovation of aggregate dividend yields (IDIV) seems totally captured by the combination
of market beta and size factor. FF factors might have played a limited role in capturing
alternative investment opportunities proxied by innovations of the selected four state
variables. We may conclude that the innovations of the selected four state variables are not
an essential element when we try to understand the average return on Chinese stock market.
Thus, we propose to consider other economic variables in this case or trying to find other
explanations of the success of FF factors in China.
Write between chapter 2 and chapter 3
The study in chapter 2 is based on one of the three specific theories that are most discussed
– exposure to changes in economic variables in the context of ICAPM (the other two
specific theories are distress risk, and asymmetric exposure to economic conditions). The
three theories are not an exhaustive list of specific theoretical explanations for the
performance of the FF3F Model. It represents three prominent theories that have empirical
support.
As the results of chapter 2 suggested that FF factors do not lose their explanatory power in
the presence of innovations of the four selected state variables (aggregate dividend yield,
one-month T-bill rate, term spread and default spread) on CNAS stock market during the
period December 2006 to May 2015. In other words, FF factors might not proxy for
innovations of selected variables on CNAS stock market over the sample period.
In this case, we seek another theoretical explanation in the following chapter 3, which is the
distress risk. Since one explanation for the persistent returns performance of high book-to-
market stocks is the risk of financial distress, which can be found in an extensive literature.
For instance, Chan and Chen (1991) argued that distressed firms are more sensitive to
changes in economic conditions and documented that distressed firms, as proxied by
dividend reductions and leverage, earned relatively high returns. Thus, they were able to
provide an explanation as to why small firms earn high returns – these firms were more
likely to have experienced dividend reductions and be highly leveraged. FF (1992)
themselves showed that stocks with high book-to-market equity ratios earned relatively
114
high returns, they proposed that this could be due to the decrease in market value associated
with lower earnings prospects for those firms in distress. Furthermore, FF and other
researchers continued to attribute the empirical evidence to distress risks in subsequent
papers (Fama and French (1995; and Fama and French (1996)).
In addition, Firm’s distress risk is an important indicator of a firm’s performance and is

also one of the most concerned characteristics by investors and firms. Measuring distress
risk has been a hot research direction since the early 1930s FitzPatrick (1932). And two
kinds of dominant models: accounting-based models that based on accounting ratios and
market-based models that based on the market information are well developed and widely
used in the massive of literature. The studies based on those predictive models are applied
for investigating the relationship between distress risk and stock returns or comparing
among the models. For instance, Dichev (1998), Griffin and Lemmon (2002), Vassalou and
Xing (2004), Chava and Purnanandam (2010). all provide evidence that stock returns are
related to distress risk (default risk).
One test of whether distress can explain why FF factors are priced risk factors was
performed by Vassalou and Xing (2004). The researchers measured the default risk of
individual stocks using a model developed by Merton (1974). Note that the risk of default
is a relatively more extreme outcome of the risk of distress. The Merton’s measure of
default risk is the basis behind the credit ratings of Moody’s KMV.
We carry out the following research and construct the distress risk factor using data of
CNAS stock market during July 2005 to May 2015. To investigate whether FF factors
proxy for distress risk on Chinese stock market, we augment FF3F Model with a mimicking
distress risk factor. Furthermore, we implement both accounting-based and market-based
model in estimating firms’ distress risk to examine whether the different methods have
effect on the empirical results.
115
3 Distress Risk Factor and Stock Returns on
Chinese A-Share Stock Market
3.1 Financial distress risk........................................................................................... 117

3.2 Measurement of financial distress risk ................................................................ 119
3.3 Financial distress risk and equity returns ............................................................. 135
3.4 Construct distress risk factor and portfolios on Chinese stock market ................ 141
3.5 Distress risk, size and B/P ratio of Chinese stock market ................................... 146
3.6 Augmented four-factor model and empirical evidence ....................................... 155
3.7 Conclusions.......................................................................................................... 167
This chapter investigates the relationship between stock returns and

distress risk, examine whether the size and value effects are related
to distress risk on Chinese A-share stock market. Furthermore, we
explore an augmented four-factor model by adding a distress risk
factor in addition to Fama-French Three-Factor Model in order to
examine whether Fama-French factors proxy for the distress risk.
To measure the financial distress risk and construct distress risk
factor, we apply both accounting-based model and market-based
model, the comparisons are performed between the results obtained
from the two different methods. Then Fama-MacBeth two-stage
approach and Errors-in-Variables adjusted method are
implemented to perform the regressions.
117
3.1 Financial distress risk
Financial distress, also known as default risk, financial crisis or financial failure, is the
situation that a company has certain kind of financial difficulties, generally, it is defined as
“the likelihood that a levered firm will not be able to pay the contractual interest or
principal on its debt obligations” Garlappi et al. (2008). The worst situation of financial
distress is bankruptcy, we say that the firm has gone into bankruptcy when a company
reaches the point where it is unable to pay its debt and must stop its economic activity.
Beaver (1966) defines the firm’s failure as the inability of a firm to pay its obligations as
they mature and points out that financial distress occurs when any of the following events
comes up: bankruptcy, bond default, an overdrawn bank account, or nonpayment of a
preferred stock dividend.
Ross et al. (1998) summarized previous studies and concluded that financial distress
contains four conditions: (1) business failure, that is, a company cannot pay the outstanding
debt after liquidation; (2) legal bankruptcy, namely, a company or its creditors applies to
the court for a declaration of bankruptcy; (3) technical bankruptcy, namely, a company
cannot fulfill the contract on schedule to repay principal and interest; and (4) accounting
bankruptcy, namely: a company’s book net assets are negative29.
While in China and some other developing countries, financial distress is usually defined as
the certain degree of financial deterioration ruled by the national security management
institution. For example, in China, a listed firm is defined as ‘Special Treatment’ (ST) by
China Securities Regulatory Commission if (1) a listed firm has negative net profits for two
years consecutively, (2) the shareholders’ equity of the company is lower than the
registered capital, and (3) a firm’s operations have stopped and there is no hope of restoring
operations in the next three months due to natural disasters, serious accidents or law-suit
and arbitration. According to the Chinese regulation, if an ST firm cannot improve its
performance within the next three years, it is labeled as ‘Particular Transfer’ (PT) and may
be delisted from the stock exchange market30.
None of the shareholders, creditors or investors is willing to witness the fail (bankruptcy) or
deterioration of the firm since this situation make them suffer severe financial losses. So
identifying firms that likely go into the deterioration situation in advance is of consequence.
In China, firm’s default risk is also one of the most concerned characteristics by investors
and firms, many policymakers and financial institutions needs to improve their
understanding of distress risk of Chinese firms. It is important to develop an early warning
29
Sun et al. (2014)
30
Geng et al. (2015)
118
3.1 Financial distress risk
system for prediction of firms’ financial distress, which have been a hot topic over the years
in China.
3.2 Measurement of financial distress risk
The primary question of predicting distress risk is how to measure it, and the financial
distress prediction, or called bankruptcy prediction, acts as an important role in the
decision-making of various areas, including accounting, finance, business, and engineering
etc. The prediction of distress risk has experienced from the qualitative analysis to
quantitative analysis, and along with the development of computer technology, the ANNA
(Artificial Nerve Network Analysis) Model which based on the statistics method and
computer, seems to be the new generation of a prediction model for distress risk. Our
research focuses on the quantitative models of measuring distress risk, which is still the
dominant methods in the research field of financial distress risk.
Early in 1932, FitzPatrick (1932) analyzes and compare the financial ratios of successful
industrial firms and those of failed firms, and he finds the significant difference between
the financial ratios of the two kinds firms. In addition, the author also points out that
financial ratios can not only reflect the financial condition but also the business
performance of a firm, what’s more, the financial ratio has predict ability for the firm’s
future. Thereafter, among
Table 3.1 Classification of financial distress prediction models
Classification Author Method/Model
Beaver (1966) Univariate Discriminant Analysis

Altman (1968) Multiple Discriminant Analysis
Accounting-based models
Ohlson (1980) Logit model
Zmijewski (1984) Probit model
Market-based models Merton (1974) Option-pricing model

(Option-pricing theory) Crosbie and Bohn (2003) KMV model
Market-based model
Shumway (2001) Hazard model
(dynamic reduced-form model)
119
numerous researches devote to predicting distress risk, the use of accounting ratios or
market information are two dominant classifications. Later in recent years, there are models
which consider both accounting and market information. Table 3.1 presents the dominating
models in predicting financial distress since 1930’.
3.2.1 Accounting-based models
The accounting-based information is an important indicator of whether a company may

encounter the financial distress or not, it can reflect a firm’s financial conditions and
business performance, and thus predict the future of the company. The early models to
predict distress risk is mainly use accounting ratios.
The first study using a statistical approach to measure distress risk based on accounting
information is Beaver (1966) univariate analysis of financial ratios which aims at predicting
the corporate failure. Six groups31 30 ratios are selected and tested on 79 failed firms and
79 sound firms during 1954 to 1964. He performs a dichotomous classification test of the
predictive ability of the chosen accounting measures and identifies the six most powerful
ratios: cash flow to total debt, net income to total assets, total debt to total assets, working
capital to total assets, current ratio, and no-credit interval. He found that those indicators
could discriminate between matched samples of failed and non-failed firms for as long as
five years prior to failure. Beaver initiates ‘The Univariate Discriminant Model’ which
make it easier for predicting finance distress using a simple model.
Despite the prominent step forward of Beaver in measuring and predicting default risk, the
univariate model cannot comprehensively predict the distress risk of a firm, for instance, (1)
single ratios calculated by Beaver do not capture time variation of financial ratios; (2)
different accounting ratios may have different predicting ability and result in different
consequences for the same firm; (3) there are interaction effects among different accounting
ratios, single ratio is not able to capture multidimensional interrelationships within the firm.
So that the interpretation of a single ratio in isolation may be incorrect. The weaknesses of
Beaver’s univariate model have led to the development of the multiple discriminant
analysis which will be the subject of the following section.
31
Cash-flow ratios, net-income ratios, debt to total-asset ratios, liquid-asset to total-asset ratios,
liquid-asset to current debt ratios and turnover ratios
120
3.2.1.1 Altman’s Z-score
To improve the accuracy of the assessment of distress risk from the univariate analysis,
Altman (1968) apply a multiple discriminant analysis (MDA) to examine 66 manufacturing
companies in the U.S., half of which filed bankruptcy while half of which are solvent
between 1946 and 1965.
Similar as Beaver, 22 accounting ratios are selected on the basis of their popularity in the
literature and their potential relevancy to the study, and are classified into five categories:
liquidity, profitability, leverage, solvency, and activity. Finally, five variables are selected
as doing the best job together to predict the corporate bankruptcy after applying the
following processes:
(1) Observation of the statistical significance of various alternative functions including

determination of the relative contributions of each independent variable; (2) evaluation of
inter-correlations between the relevant variables; (3) observation of the predictive
accuracy of the various profiles; and (4) judgment of the analyst.
The five variables constitute the final discriminant function, which we call it Z-score:
Z  0.012 X 1  0.014 X 2  0.033X 3  0.006 X 4  0.999 X 5 (3.1)
Where, Z is overall index (Z-score),

X 1 is Working capital/Total assets,
X 2 is Retained Earnings/Total assets,
X 3 is Earnings before interest and taxes/Total assets,
X 4 is Market value equity/Book value of total debts,
X 5 is Sales/Total assets.
The lower a firm’s Z-Score, the higher its default probability (DP). Altman proposes the
“cut-off” point (critical point) so that to predict at what level to bankruptcy a firm is
according to its Z-score. It is concluded that firms which have a Z-score below 1.81 are all
bankrupt; the firms having Z-score between 1.81 to 2.675 are defined in a “gray area” or
“zone of ignorance”, which means the firms are in a situation that is not so clear, they have
the probability to bankrupt; while firms having a Z-score bigger than 2.99 clearly are in the
non-bankrupt condition. Still, it is uncertain about a firm whose Z-score is in the “zone of
ignorance”, Altman obtains a critical Z value 2.675, which is to tell from the bankrupt and
non-bankrupt firms in this area.
121
Altman’s Z-score improves accounting-based techniques of the identification of financial

distress risk, and it has advantages over Beaver’s univariate model by considers a set of
weighted combined accounting ratios. Altman proves that his model has an extremely
accurate in predicting bankruptcy.
However, MDA has some well-known limitations. Such as (1) the statistical assumptions
that predictors need to be normally distributed and the variance-covariance matrices of the
predictors should be the same for both groups of firms (bankrupt and non-bankrupt); (2) the
output of MDA model is a value (or score) which has little intuitive interpretation, since it
is basically an ordinal ranking (discriminatory) device. (3) the accuracy of Altman’s model
two or three year prior to default drops drastically than the accuracy one year prior to
default, the reason maybe that deteriorate of some accounting ratios is just the appearance
instead of the essence of bankruptcy.
3.2.1.2 Ohlson’s O-score
Ohlson (1980) criticizes the restrictive assumptions of Altman’s MDA and comment that
previous studies appear to have overstated the predictive power of models. To avoid the
problem of MDA, Ohlson is the first who introduces the conditional logit analysis which is
one of the conditional probability analysis to predict the probability of default and estimate
firms’ failure. The major advantage of the logit analysis is that ‘no assumptions have to be
made regarding prior probabilities of bankruptcy and/or the distribution of predictors’.
The logit model is based on the cumulative distribution function to maximize the joint
probability of default for the distressed firms and the probability of non-failure for the
healthy companies in the sample. Similar to the MDA, this method weights the independent
variables and assigns a score, however, this method estimates the probabilities of default
for each company in a sample.
Suppose P as the probability of default for any given firm, let X i denote the predictors for
the i th observation, let Wi be the parameters of X i , the expression of logit model is:
1 1
P (  yi )
 ( W0 W1 X 1 W2 X 2  Wn X n )
(3.2)
1 e 1 e
Doing a little transformation and we will have:

n n
P
yi  ln( )  W0  W1 X 1  W2 X 2   Wn X n   Wi   X i (3.3)
1 P i 0 i 1
122
The logit model supposes that ln( P / ( 1  P )) can be explained linearly by accounting
ratios. P is the probability function which value is between 0 and 1 ( 0  P  1 )32, the cutoff
is 0.5, which means that if P bigger than 0.5, companies tend to have more probability of
default, otherwise, the companies are healthy. Unlike the MDA which use a Z-score value
to predict whether a firm is bankrupt or not, logit model uses a probability to measure the
default risk, which makes it more accurate and reliable.
Ohlson chooses the sample that contains 105 firms which are bankrupt and 2058 non-
bankrupt firms between the period 1970 and 1976, and all the firms are classified as an
industrial. He constructs three logit models, model 1 predicts bankruptcy within one year,
model 2 predicts bankruptcy within two years, and model 3 predicts bankruptcy within one
or two years. The statistic “Percent Correctly Predicted” of the three models are 96.12%,
95.55%, 92.84% respectively.
He identifies four factors: the size of the company, a measure(s) of the financial structure, a
measure(s) of performance and a measure(s) of current liquidity; which are statistically
significant in affecting the probability of firm failure.
Finally, nine independent variables are employed to determine the probability of

bankruptcy:
- SIZE = log (total assets / GNP price-level index)

- TLTA = total liabilities / total assets
- WCTA = working capital / total assets
- CLCA = current liabilities / current assets
- OENEG = 1 if total liabilities > total assets, 0 if otherwise
- NITA = net income / total assets
- FUTL = funds from operations / total liabilities
- INTWO = 1 if a net loss for the last two years, 0 otherwise
- CHIN = ( net income t - net income t 1 ) / ( net incomet  net income t 1 )
Assigning the corresponding weights to each variable and get the final O-score formula:
O  1.32  0.407 log( SIZE )  6.03(TLTA)  1.43(WCTA)  0.076 (CLCA)

(3.4)
1.72(OENEG )  2.37 ( NITA)  1.83( FUTL)  0.285( INTWO)  0.521(CHIN )
In which, two of them are dummies (OENEG and INTWO). The use of qualitative
variables is another advantage of the logit model compared to the discriminant analysis
which is limited to the interpretation of quantitative ratios.
32
Refer to McFadden and others (1973) for a comprehensive analysis of the logit model.
123
Contrary to Altman’s Z-score, the lower a firm’s Z-Score, the more likely of a firm
encounter default, the probability of default positively changes with Ohlson’s O-score,
which means the higher the O-score the higher the default risk.
Ohlson’s logit model have several advantages: the predictors do not need to follow the
normal distribution or the same variance-covariance matrices rule, and unlike Altman’s
MDA of which the output is a value, the logit model is based on the conditional probability
analysis and measures the firm’s probability of default. come to the conclusion that the
logit model generally is superior to the MDA approach of Altman. Kleinert (2014)
compares the performance of three accounting-based models in Belgium and Germany and
draw the conclusion that Ohlson’s logit model performs most accurate. In the Asian market,
Wang and Campbell (2010a) studied listed Chinese companies during 2000 to 2008 and
report a high accuracy rate (95%) of Ohlson’s model. Pongsatat et al. (2004) conclude that
the Ohlson’s logit model has a higher predictive ability in all three years preceding
bankruptcy than that of Altman’s MDA approach in Thailand.
Despite all the advantages, some critics are left on Ohlson’s logit model: the use of
maximum likelihood methodology to estimate the parameters makes the computational
procedure complex. Hillegeist (2004) argues that Ohlson’s logit model fails by not
including time varying changes. Grice and Dugan (2001) emphasizes that the relation
between financial ratios and their effects on bankruptcy changes over industries and time.
Ohlson himself gives advice that the choice of different accounting ratios could improve
the likelihood function. However, he also suggests that such non-accounting information as
equity prices or their volatility might be most useful and should be examined in future
research. The use of non-accounting information for predicting financial distress has led to
the development of a special class of default-risk models based on the value of a firm set by
the market.
3.2.2 Market-based models
Since the prevalent of accounting-based models in modeling the default risk, there follows
some criticism about the decline performance and limitation of using accounting variables
to predict distress risk33.
33
Lev and Zarowin (1999), Mensah (1984), Hillegeist et al. (2004), Wu et al. (2010), Begley et al.
(1996).
124
Several reasons are listed below to query the distress risk measures which are based on
accounting data:
- Default probability is a prediction about the likelihood of future events, though

accounting information is backward-looking. Accounting models use information
derived from the financial statements which aim to measure past performance and
may not be very informative in predicting the future status of the firm. As Hillegeist
et al. (2004) argue that since the accounting statements are prepared on a going-
concern basis, they are of limited utility in predicting bankruptcy by design.
- In addition, “the conservatism principle often causes asset values to be understated

relative to their market values, particularly for fixed assets and intangibles.
Downward-biased asset valuations will cause accounting-based leverage measures
to be overstated.” Hillegeist et al. (2004)
- Financial ratios vary substantially across industries. Thus, accounting-based

coefficients are specific to the industry and sample used and cannot be generalized
with respect to all firms in the market.
- Accounting-ratio-based models are typically built by searching through a large

number of accounting ratios with the ratio weightings estimated on a sample of
failed and non-failed firms. Since the ratios and their weightings are derived from
sample analysis, such models are likely to be sample specific. An additional point of
critics has been that accounting models ignore economic idiosyncrasies and that
data are collected over many years while leaving out market changes Mensah
(1984).
- Most importantly, another deficiency is that accounting-based models do not

incorporate a measure of asset volatility. Volatility is a crucial variable in analyzing
and predicting bankruptcy because it captures the likelihood that the value of the
firm’s assets will decline to such an extent that the firm will be unable to repay its
debts.
Those limitations of models using accounting variables in modeling default risk bring on
the models that rely on market information. The equity market contains an alternative and
potentially superior source of information about default probability because it assembles
information from other sources in addition to the financial statements.
There are two classes of models that based on market data, specifically, structural models
Merton (1974) use option pricing methods to compute the default probability from the level
125
and volatility of market value of assets, and reduced-form models34 Shumway (2001) allow
the default intensity to be extracted from debt or credit market securities. Since the vast
majority of the market-based models are on the basis of option-pricing theory, we will
mainly introduce the two popular model derived from option pricing theory and a model
which belongs to the class of dynamic reduced-form models of default.
3.2.2.1 Black-Scholes and Merton (BSM) option-pricing model
Merton (1974) is the first who proposed a market-based model which applies the option-
pricing methodology developed by Black and Scholes (1973) to relates the default risk to
the capital structure of the company. In Merton’s model, the equity (common stock) of a
firm can be viewed as a standard European call option on the underlying firm’s assets with
a strike price equal to the book value of the firm’s liabilities.
The limited liability feature of equity means that the equity holders have the right, but not
the obligation, to pay off the debt holders and take over the remaining assets of the firm.
That is, the holders of the other liabilities of the firm essentially own the firm until those
liabilities are paid off in full by the equity holders. Thus, in the simplest case, equity is the
same as a call option on the firm’s assets with a strike price equal to the book value of the
firm’s liabilities Crosbie and Bohn (2003).
The reason is that shareholders are residual claimants on the firm’s assets after all other
obligations have been met. According to Merton’s theory, it is just as the shareholders sold
the corporation to their creditors, they have the right, but not the obligation to pay off the
creditors. The relation between equity value VE (y-axis) and firm’s asset value V A (x-axis)
is shown in Figure 3.1. At the maturity of the option, if the value of the firm’s assets (take
the point V2 as example) is greater than the book value of liabilities X, the shareholders
exercise their option on the assets, and the firm continues to exist. Otherwise, if the firm’s
asset value (take the point V1 as example) is lower than the book value of liability X at
maturity, the shareholders will choose not to exercise the option right and the value of
equity is zero, which means the firm will default.
Thus, the market value and volatility of the firm’s underlying assets implied by the equity’s
market value are important to determine to what extent a firm will go bankruptcy. However,
the market value and volatility of the firm’s assets are usually cannot obtained directly. In
34
Jarrow and Turnbull (1995), Duffie and Singleton (1999)
126
particular, Merton solves backward from the option price and option price volatility for the
implied asset value and asset volatility.
Figure 3.1 Equity as a European call option on the firm
Firstly, recall that the assumption of Black-Scholes (BS) model is that the market value of a
firm’s underlying assets follows the following Geometric Brownian Motion (GBM):
dVA  VA dt   AVA dW (3.5)
where V A and dVA are the firm’s asset value and its change,  and  A are respectively an
instantaneous drift rate and the instantaneous volatility of firm’s asset value, and dW is a
standard Weiner process.
Besides, the BS model also assumes that the capital structure has only a single class of debt
and a single class of equity.
Then, under these assumption and following Merton’s theory, denote X as the book value of
liability that maturing at T, the market value of firm’s equity VE can be seen as a European
call option on firm’s underlying assets V A with has maturity equal to T, which is given by
the BS formula for call options:
VE  VA N  d1   Xe rT N  d 2  (3.6)
where,
127
 1   1 
ln VA X    r   A2  T ln VA X    r-  A2  T
d 1=  2 
, d 2  d1   A T   2  ,
A T A T
N is the cumulative density function of the standard normal distribution, r is the risk-free
rate, and Xe  rT is the present value of the promised debt payment.
Next, to solve the two unknown parameters V A and  A in equation (3.6) which implies that
the equity value can be represented as a function of the asset value, Merton applies Ito's
Lemma35 to determine an instantaneous standard deviation of equity that can be otherwise
estimated from the historical share prices.
Follows from Ito’s lemma, the relation between equity value volatility  E and the asset
value volatility  A is as follows:
VA V V V
E   A  A  E   A  A  N  d1    A (3.7)
VE VE VA VE
 VE
where the hedge ratio  equals to which is the partial derivative of the equity value
V A
with respect to the asset value, and from BS formula of equation (3.6), it can be shown that
 VE
the hedge ratio = N  d1  .
V A
Combines equation (3.6) and equation (3.7), asset value and its volatility are calculated.
The default probability DPt is the probability that the firm’s assets value is less than the
book value of the firm’s liabilities, based on which, Merton derives the firm’s probability of
default (the details of derivation process are shown in Appendix E) in terms of the
cumulative normal distribution:
  VA,t   1 2 
 ln        A T 
DP=N    Xt   2   (3.8)
t
 A T 
 
 
35
For a rigorous discussion of Ito's Lemma, see McKean (1969) and for references to its application
in portfolio theory, see Merton (1973b)
128
Merton’s distance-to-default (DD) is the number of standard deviations that the firm is
away from default, which means that the higher the DD, the farther the firm is away from
default (the lower probability of default), on the contrary, the lower the DD, the higher
probability of the firm bankrupt.
 1 
ln VA,t / X t       A2  T
DD   2 
(3.9)
A T
DP  N   DD  (3.10)
The major advantage of BSM option-pricing model in predicting default risk is that “they
provide guidance about the theoretical determinants of bankruptcy risk and they supply the
necessary structure to extract bankruptcy-related information” Hillegeist et al. (2004).
Unlike the accounting-based models that are constructed by comparing the characteristics
of bankrupt and no bankrupt firms, using a statistical technique to derive the variables that
best discriminate between the two groups of firms which are not grounded in theory, and
distinguish firms of bankruptcy or no bankruptcy heavily dependent on the prior
specification of firms. The equity of a firm can be viewed as a call option on the firm’s
assets leads to a measure of default risk that is derived from theory and is economically
justifiable.
Furthermore, compared with the conventional accounting-based models, it is more forward-

looking, dynamic and easier for using because it is calculated based on the market prices
which reflect future expected cash flows (the accounting-based models, on the contrary,
reflect the past performance of the firms), and contain the information comes from financial
statements plus other information not included in the financial statements.
Among plenty researches that implement BSM model, one of the most popular and
commonly used is KMV model, which will present in the next section.
3.2.2.2 KMV model
KMV model was first introduced by Oldrich Vasicek and Stephen Kealhofer Vasicek (1984)
also as an extension of BSM model. In 1989, Kealhofer, McQuown, and Vasicek set up the
KMV Company which is named by the founder’s names and later in 2002 was bought by
the company Moody´s. Then KMV is set to the successful practical model that estimated
the default risk of firms.
129
Though BSM option-pricing model is the genesis for understanding the link between the
market value of the firm’s assets and its equity, and is widely understood and provides a
useful framework to estimate firm’s DP. KMV model assumes “the firm’s equity is a
perpetual option with the default point acting as the absorbing barrier for the firm’s asset
value” Crosbie and Bohn (2003), the firm is assumed to default when the asset value hits
the default point. Unlike BSM model, allows only two types of liabilities, a single class of
debt and a single class of equity, KMV model takes multiple classes including short-term
liabilities, long-term liabilities, convertible debt, preferred equity, and common equity of
liabilities into consideration.
There are three main steps to determine default probability of KMV model, we will present
precisely the differences from BSM model in each step.
- Step 1 Estimate firm’s asset value and its volatility

The BSM model applies the method called “simultaneous equations method” to solve
firm’s asset value and volatility, that is solving the combination of tow equations (3.6) and
(3.7) to obtain two unknown parameters. However, the model which presents the links
between asset value and volatility given by equation (3.7) holds instantaneously. (Crosbie
and Bohn, 2003) emphasize that “In practice the market leverage moves around far too
much for equation (3.7) to provide reasonable results. Worse yet, the model biases the
probabilities in precisely the wrong direction”. For instance, in equation (3.7), if the market
leverage ( VA / VE ) decreases then the asset volatility  E will tend to be overestimated and
thus the default probability will be overstated as the firm’s credit risk improves. On the
conversely, if the market leverage increases the asset volatility  E will be underestimated
and the default probability will be understated.
Instead of using instantaneous relationship in equation (3.7) and “simultaneous equations

method” to solve the two unknown parameters of the Merton' option pricing formula (3.6),
KMV model Crosbie and Bohn (2003) resolve the problem by using a complex iterative
procedure to find value of firm’s asset and volatility. The procedure uses an initial guess of
firm’s asset volatility  A and then solve the equation (3.6) to determine a set of the firm’s
asset value V A and thus obtain the asset returns. Then the volatility of the resulting returns
is used for the next iteration to determine a new set of V A and therefore a new series of
asset returns. The procedure continues like this until the results converge.
- Step 2 Calculate the default point and distance-to-default

To calculate the DD of Moody’s KMV, an important concept is default point. A firm
default if the market value of assets falls below a certain value which is called the default
130
point, and it is generally accepted that the firm reaches the default point when the value of
asset less than its total liabilities. However, Crosbie and Bohn (2003) find in general this is
not the case, the default commonly occurs when the market value of firm’s asset locates
between the total liabilities or total debt and the short-term (current) liabilities (debt). In
KMV model, the default point is calculated as short-term debts (STD) plus one-half of
long-term debts (LTD).
1
Default Point= STD+ LTD (3.11)
2
Moody’s KMV argues that in fact, the distribution of the DD is difficult to measure, so the
assumption of normal or lognormal distribution such as BSM model is not appropriate to
use in practice. KMV model proposes the DD which “compares the market net worth to the
size of a one standard deviation move in the asset value”:
VA  Default Point
DDKMV = (3.12)
VA A
where V A is the market value of firm’s assets and  A is asset volatility. The numerator in
the formula is the firm’s market net worth which equals to the value of assets minus the
default point, a firm will default when its market net worth reaches zero.
- Step 3 Calculate the Expected Default Frequency based on the empirical

distribution of the DD
BSM model calculates the DP (default probability) as the normal distribution of negative
DD, whereas, Crosbie and Bohn (2003) point out that in practice the normal distribution is
a very poor choice to define the probability of default, because Moody’s-KMV observe that
defaulted firms have a leptokurtic distribution and the normal distribution underlying the
BSM model leads to underestimation of the true value of firm’s DP.
To avoid this effect of using the normal distribution, after calculating the distance-to-
default, the KMV use their own large default database36 which collected over 20 years, to
derive an empirical distribution relating the DD to the DP. The empirical distribution
obtained from the KMV’s proprietary database has much wider tails than the normal
distribution.
This DP is well known as Expected Default Frequency (EDF), which is the market-based
credit measure developed by Moody’s KMV. The EDF is nothing but the probability that a
36
Their database includes over 250,000 company-years of data and over 4,700 incidents of default
or bankruptcy.
131
given firm will default within 1 year according to the KMV methodology. The formula of
EDF is similar as BSM’s DP of equation (3.10), the normal distribution N is replaced by
the empirical distribution of KMV which we can denote ÑKMV :
EDF = ÑKMV  DDKMV  (3.13)
Moody’s KMV has also a software product which is called ‘Credit Monitor’ to analyze the
EDF credit measures, one can calculate the EDF values for one to five years through Credit
Monitor.
Many researchers go for KMV’s claim and argue that it is inconsistent to derive a formula
for calculating DP based on an underlying normal distribution and then depart from KMV’s
empirical distribution in the final calculation of the DP. A large number of world financial
institutions are subscribers of the KMV model.
From a purely theoretical point of view, the differences between KMV and Merton’s
models are not dramatic. The KMV model, however, relies on an extensive empirical
testing and it is implemented using a very large proprietary database.
3.2.3 Hazard model
Shumway (2001) find that many accounting ratios used in previous models (Altman (1968
and Zmijewski (1984) for estimating default probability are not statistically significant,
instead, the market size, past stock returns and idiosyncratic standard deviation of stock
returns are all strongly related to the firm’s bankruptcy. The author argues that the static
models do not consider the time-changing characteristics of firms and propose a discrete-
time hazard model to predict bankruptcy by using both accounting ratios and market
variables, which is a better predictor of bankruptcy than alternative models.
The concept of Shumway’s hazard model comes from the survival model that estimates
firm’s hazard rate according to the sample’s survival condition. In other words, “a typical
discrete-time hazard at time t can be interpreted as a conditional probability of default at
time t, given that the default did not happen prior to time t Outecheva (2007)”. There are
two main functions to understand the hazard model: the survivor function
S  t,x;   1   f  j,x;  (3.14)

j t
and the hazard function
132
f  t,x; 
  t,x;   (3.15)
S  t,x; 
where, f  t,x;  is the probability function of default, and θ is the vector of parameters of f,
x is a vector of explanatory variables for predicting default. The survivor function (3.14)
gives the probability of surviving up to time t, and the hazard function (3.15) gives the
probability of failure at t conditional on surviving to t
To estimate hazard function, many of which are difficult to estimate because of their
nonlinear likelihood functions and time-varying covariates. Shumway shows that the
discrete-time hazard model has the same likelihood function as logit model:
n
L=   ti ,xi ;  i S  ti ,xi ; 
y
(3.16)
i 1
yi is a dummy variable which equals one if firm i default at ti or equal to zero if otherwise.
So it is possible to estimate a hazard model with a logit program by “adjusting the sample
size assumed by the logit program to account for the lack of independence between firm-
year observations”37.
Chava and Jarrow (2004) prove that the prediction ability for the bankruptcy of Shumway’s
hazard model is superior to accounting-based models (Altman (1968) and Zmijewski
(1984)), in addition, they find that “accounting variables add little predictive power when
market variables are already included in the bankruptcy model”. Campbell et al. (2008)
also develop a dynamic hazard model to estimate the default probabilities and study the
determinants of corporate bankruptcy and the pricing of distressed stocks. They declare that
their best model includes additional variables and has greater explanatory power than the
existing models estimated by Shumway (2001) and Chava and Jarrow (2004).
3.2.4 Comparing Accounting-based models and market-based models
In summary, the market-based models overcome the main shortcomings we mentioned at

the beginning of the section 3.2.2 about accounting-based models:
37
In the static logit model, the number of firm-years is used in calculating the Wald statistics.
However, this is not correct for the dynamic logit model because in the dynamic logit model, unlike
the static logit model, firm-year observations are not independent of each other. For the dynamic
logit model, it is the number of firms rather than the number of firm-years that should be used.
133
- The most importantly, market-based models employ BS option-pricing theory

provide the theoretical basis for predicting firm’s default risk or distress risk. While
the accounting-based models are not grounded on the economic theory.
- Dissimilar accounting-based models that use information derived from financial

statements which reflect the past performance of the firms, market-based model
such as BSM model, depend on the market prices which reflect investors’
expectations about the firm’s future performance, thus the market-based models are
better for calculating the probability of default in the future. Furthermore, the
market prices contain the information in financial statements and other information
which is not included in the financial statement.
- Market-based variables provide the direct estimation of volatility, which is a

powerful predictor of firm’s default risk and is not contained in the accounting-
based model.
Hillegeist et al. (2004) compare the performance of two accounting-based models

(Altman’s Z-score and Ohlson’s O-score) and the market-based model (BSM option-
pricing model), and their results demonstrate that BSM model provides significantly more
information of bankruptcy probability than both accounting-based models. They
recommend researchers to apply BSM model instead of Z-score and O-score proceeding the
research.
Despite the popularity and the advantages of market-based models to predict firm’s default
risk, there are still some critics left for market-based models, such as the models based on
the underlying option-pricing theory require the assumption of normality of stock returns
and the models do not distinguish between different types of debt and assume that the firms
only have a single zero coupon loan38. Many researchers give the evidence that accounting-
based models are still indispensable in predicting distress risk.
Using a hazard model with a longer time period, Beaver et al. (2005) finds that the ability
of accounting ratios to predict bankruptcy remains. Their findings indicate that the market
variables complement accounting variables and that the use of market variables causes only
a slight reduction of predictive power of the accounting variables in certain sub-periods.
Agarwal and Taffler (2008) find out that there is little difference between market-based
models and Altman’s model in predicting firms’ distress risk in the UK, and the
accounting-based model even produces significant economic benefit over the market-based
models. In the paper of Reisz and Perlich (2007), they estimate the default probability for
38
Allen and Saunders (2002)
134
5784 industrial firms and conclude that Altman’s Z-score model better predict firm’s
bankruptcy over a one-year horizon. Moreover, Campbell et al. (2008) find that market-
based models have little forecasting power after controlling for other variables.
To conclude, when comparing the two mainstream distress risk prediction models:
accounting-based models and market–based models, we can say that both imply advantages
and disadvantages.
3.3 Financial distress risk and equity returns
In the asset pricing area, as we all know that the market risk cannot explain all the variation
of average stock returns, researchers have been committed to exploit the anomalies which
have additional explanatory power. One of the anomalies that researchers are interested in
is the firm’s distress risk, the cross-sectional relation between default risk and stock returns,
the so-called default risk premium, has been a subject of intense debate in the literature.
numerous studies apply the models for measuring distress risk and try to find the relation
between expected stock returns and firm’s distress risk.
Since a number of empirical studies investigating the relationship between financial distress
risk and equity returns find that the distress risk is related to firm’s B/M ratio, especially
after FF3F model achieves huge empirical success, considerable researchers regard
financial distress risk as a factor and examine whether distress risk factor can be explained
(proxy) by FF factors.
3.3.1 Literature reviews on developed markets
3.3.1.1 Distress risk and expected stock returns
In order to find out the relation between distress risk and equity returns, it is important to
have a correct proxy for distress risk, and “the risk of bankruptcy appears to be one natural
measure of firm distress” Dichev (1998), thus the diverse models of measuring firm’s
default probability provide the rational proxies for distress risk. The empirical studies on
the distress premium mainly apply either the traditional accounting-based models such as
Altman’s Z-score and Ohlson’s O-score39 or the market-based models such as Merton’s
39
Dichev (1998), Griffin and Lemmon (2002) and Ferguson and Shockley (2003), etc.
135
option-pricing model and Moody’s KMV model 40 to estimate distress risk (default
risk/probability). A few researchers also apply the dynamic reduced-form model such as
Shumway’s hazard model as a proxy for distress risk.
Dichev (1998) is the first who carries out researches on the relationship between default
risk and stock returns, by applying Altman’s Z-score and Ohlson’s O-score as proxies for
default risk. Dichev examines whether bankruptcy risk is a systematic risk and the relation
of distress risk factor to size and B/M effects on U.S. market during the period 1981 to
1995. His findings reveal that the risk of bankruptcy is not a systematic risk and there is a
negative relationship between bankruptcy risk and stock returns, firms with higher default
risk are rewarded by lower average returns instead of higher returns, which is inconsistent
with the risk-based explanation for distress, this puzzling relation between distress risk and
stock returns is often called the “distress anomaly. Furthermore, inconsistent with some
studies41 which suggest that the size and B/M effects may the proxies of the distress risk
factor, his results demonstrate that the relationship between bankruptcy risk and B/M equity
ratio is not monotonic, thus “a return premium related to bankruptcy risk cannot fully
explain the B/M effect”, and there is no size effect during his research period.
Dichev’s conclusion is confirmed by Griffin and Lemmon (2002) who examines the
relationship between B/M equity, distress risk and stock returns using Ohlson’s O-score as
a proxy for firm’s distress risk. Similarly, they find that firms with high B/M equity ratio
earn significantly lower returns as a premium for distress risk than those of low B/M firms
and they come to a conclusion that the B/M effect must be due to mispricing.
Numerous researches following obtain the negative relationship as Dichev, Campbell et al.
(2008) and Campbell et al. (2011) implement a dynamic hazard model for predicting firm’s
distress risk by combining both accounting information and market-based measures over
the period 1981 to 2003, and they document that the default probability is significantly
negatively related to stock returns, which becomes even more stronger after controlling for
size. They give some possible explanation for the distress anomaly: the stocks of firms with
higher distress risk are overestimated by investors, and “distressed stocks have
characteristics that appeal to certain investors, such as increased opportunities to extract
private benefits of control or positive skewness of returns”. Likewise, George and Hwang
(2010) also report a negative relation between stock returns and default risk measured by
the O-score.
However, there are a few researchers find the positive relationship between distress risk and
stock returns, such as Vassalou and Xing (2004) which is the first study that examines
40
Vassalou and Xing (2004), Garlappi and Yan (2011) Gharghori et al. (2009), etc.
41
Fama and French (1992), K. C. Chan et al. (1985), Chan and Chen (1991), etc.
136
3.4 Construct distress risk factor and portfolios on Chinese stock market
default risk in the context of the Fama-French model by extracting default risk estimated
from Merton’s option pricing model. Inconsistent with Dichev, they point out that the
default risk is a systematic risk, and the default risk is closely related to the firm size and
B/M equity ratio, there exist size and B/M effects but only within the highest default risk
portfolios. Moreover, contrary to the previous empirical studies, firms with higher default
risk earn higher returns than low default risk firms, however, this finding only lies in the
small size and high B/M portfolios. They attribute these conflicting results to the
measurement for estimating default risk.
Chava and Purnanandam (2010) also document a positive and stable cross-sectional relation
between default risk and expected stock returns throughout their research period. They
argue that the negative relation between distress risk and expected returns is due to “the use
of noisy ex-post realized return by the prior studies to estimate the ex-ante expected return”.
Then the implied cost of capital (ICC) is applied to estimate the expected stock returns, and
the two measures to predict firms’ default risk are EDF of KMV model and Shumway’s
hazard model.
Recently, apart from using accounting-based models or market-based models as proxy of

distress risk, Friewald et al. (2014) examine the relation between expected stock returns and
credit risk premium estimated from credit default swaps (CDS) spreads, as they claim that
“sorting firms into portfolios using only physical or risk-neutral default probabilities may
not be sufficiently informative about expected stock returns”. Their results indicate a
positive relationship between the expected excess stock returns and default probabilities,
the stock returns increase with credit risk premium. Their findings are conflictive with
Avramov et al. (2007) who use credit ratings to measure financial distress, show that stock
returns significantly increase with S&P senior debt credit ratings, which implies a negative
relationship between returns and default risk. show that most of the negative return for high
default risk stocks is concentrated around rating downgrades.
The literature generally has the conclusions that there exist negative or positive or no
relationship between stock returns and distress risk. Nonetheless Garlappi et al. (2008) and
Garlappi and Yan (2011) document particularly a hump-shaped relationship using market-
based default probability estimates from Moody’s KMV model. The results demonstrate
that the higher distress risk not always associated with higher expected stock returns.
Furthermore, they emphasize the important role of the ‘shareholder advantage’ in
determining the equity returns, the higher shareholder advantage the lower expected equity
returns as the default probability increases, and lower shareholder advantage which implies
a higher expected equity returns is positively related to default risk. As they expressed “the
trade-off between the risk of default to equity and the likelihood of bargaining gains in
137
renegotiation results in a hump-shaped relationship between expected returns and default

probability”.
Out of U.S. market, Gharghori et al. (2009) test the relationship between default risk and
equity returns in Australia using option-based models as proxy for distress risk. Their
results show that there is a negative relationship between default probability and stock
returns, what’s more, they also suggest that the negative relationship is not because of a
leverage, volatility or momentum effect.
Aretz et al. (2014) provide the evidence on the cross-sectional relation between default risk
and stock returns for non-U.S. firms in 14 developed markets using the approach of
Campbell et al. (2008, 2011). Their results demonstrate a statistically significant positive
default risk premium in the 14 international markets, the firms with higher default risk
outperform those with lower default risk. In addition, their results also indicate that “the
magnitude of the default risk premium is contingent upon several firm characteristics”.
3.3.1.2 Distress risk factor
Vassalou and Xing (2004) point out that their default probability (DP) estimated using
Merton’s model is not the real DP, instead, the DP calculated based on the empirical
distribution by Moody’s KMV model is the actual DP. Thus they call their measure of DP
default likelihood indicator (DLI). In order to construct the distress risk factor, they proceed
following steps: first state the aggregate default likelihood measure P(D) which is the
simple average of the DLI of all firms; then define the aggregate survival rate SV equals to
1- P(D); finally, the distress risk factor is defined as the change of SV:
  SV   SVt  SVt 1 (3.17)
Equation (3.17) represents the change in aggregate survival rate   SV  at time t. The
authors also examine the correlation between   SV  and several commonly used default
spreads 42, the low correlations indicate that the information captured by their   SV  is
different from that captured by those default spreads.
42
Default return spread, which is defined as the return difference between Moody’s BAA rating
bonds and AAA rating bonds. Default yield spread, which is defined as the yield difference between
Moody’s BAA bonds and AAA bonds. The change in default spread from Hahn and Lee (2001),
which is defined as the change of the difference in the yields between Moody’s BAA bonds and
100-year government bonds.
138
Another distress risk factor proposed by Gharghori et al. (2007) is a mimicking portfolio
which measures the difference in returns between high default risk firms and low default
risk firms. They also apply Merton’s option-pricing model to calculate firms’ default
probability, and similar as Vassalou and Xing, they augment FF3F model with their distress
risk factor to test whether FF factors SMB and HML can proxy for distress risk. We will
follow the later method to construct our distress risk factor on CNAS stock market.
3.3.2 Empirical studies of distress risk on Chinese stock market
To the best of our knowledge, the research on Chinese market mainly focuses on testing the
predictive accuracy of commonly used models in estimating distress risk using data of
Chinese listed firms and measuring default risk of Chinese companies without considering
the relationship between default risk and stock returns.
Wang and Campbell (2010b) investigate the prediction accuracy of Altman’s Z-score in
estimating Chinese firms’ failure using data from 2000 to 2008, and point out that Z-score
has high accuracy in predicting firms’ failure in China. In the same year, they Wang and
Campbell (2010a) publish an article that examines the predictive ability of another leading
accounting-based model (Ohlson’s O-score) for Chinese listed firms, in which the
prediction accuracy rate is testified above 95% in general. Likewise, Ni et al. (2014) show
that accounting-based models (Altman’s Z-score and Ohlson’s O-score) perform
reasonably well in determining business failures of Chinese firms.
Huang et al. (2013) use Altman’s Z-score and Ohlson’s O-score to measure the financial
distress risk of all nonfinancial firms of A-share stock market for the period 1997 to 2008,
trying to test whether there exist value or size effect. Their results show that firms with
small size have higher distress probability than firms with big size, while there is no
significant difference in distress probability for firms with different B/M equity ratio. In
addition, when the distress risk factor is added into their conditional regressions, the size
factor SMB of FF3F model lose more than half of explanatory power, however, “the proxy
for distress risk itself does not show incremental explanatory power when competing with
the three Fama-French factors”. Overall, they conclude that the size factor is better than
distress risk factor in explaining variations of expected stock returns on Chinese stock
market.
Apart from the literature that tests the accounting-based models in China, there are
researchers examine the practicability of the market-based model (Merton’s model and
Moody’s KMV model) in predicting distress risk of Chinese listed firms. Such as Lu et al.
(2006), Zhou and YANG (2007), Xiaohong Chen et al. (2010) and Chen and Chu (2014),
139
who apply KMV model to estimate distress risk; Liang (2012) and Law and Roache (2015),
who calculate firms’ distress risk using models based on Merton’s theory.
Particularly worth mentioning, most of the literature using KMV model to measure distress
risk of Chinese firms have discussed that it is not appropriate to apply KMV’s EDF in
calculating firms’ default probability, since KMV model use an empirical distribution
derived based on the database of U.S. market. At present, as stated by Xiaohong Chen et al.
(2010), “there is no similar database in Chinese credit market, which means no EDF
statistics is available”, so researchers use DD of KMV model instead of the DP as the
measure of distress risk in China.
Xiaohong Chen et al. (2010) measure credit risk of listed firms of Small Medium Enterprise
Board (SME) in China using KMV’s DD and evidence indicates the high credit risk of
listed firms in SME. Moreover, they find firms’ size has a significant impact on credit risk,
the small size firms tend to have higher default probability than medium and large size
firms. Chen and Chu (2014) implement an empirical research for the default risk of Chinese
real estate firms during the period 2007 to 2012. They implement KMV model as proxy for
distress risk and contrary to Xiaohong Chen et al. (2010) that they find big size firms face
higher default risk than small size firms.
Liang (2012) tests the predictive ability of DLI (Vassalou and Xing, 2004) by using data of
Chinese listed firms during 2000 to 2010, and the results reveal that the DLI based on
Merton’s model is a significant model for predicting distress risk on Chinese stock market.
What’s more, the author augments the original Merton-KMV model with financial ratios
(profitability, leverage, and liquidity) and compare the augmented model with Merton-
KMV model and accounting-based model. The evidence shows that the original Merton-
KMV model can be improved by an augmented model and the accounting-based model is
the weakest one in measuring default risk on Chinese stock market.
More recently, Law and Roache (2015) assess Chinese firms’ distress risk using a variant of
Merton’s model and link the default risk with firm-specific and economic variables to test
whether they have an influence on firms’ default probabilities. Furthermore, the authors
compare the result with that obtained from a model using borrowing cost as a measurement
of default probability. They conclude that the distress risk measured by the market-based
model is affected by firms’ fundamentals, and that market-based model better estimates the
stand-alone 1-year probability of default for individual firms in China.
Several researchers exploit distress predictive models themselves depending on the data of
Chinese firms, such as Geng et al. (2015), who build financial distress warning model
140
based on 32 financial indicators of 107 Chinese firms which are ‘special treatment’ (ST)43.
And some other researchers such as Bhattacharjee and Han (2014) and Liu and Wang (2016)
trying to find the impact of some macroeconomic, firm-specific variables and the cutoff
point on the financial distress risk of Chinese listed firms.
Lin and Chen (2008) is one of the few studies that investigate the relationship between
distress risk and expected stock returns on Chinese stock market. They examine whether
default risk is a systematic risk in China and related to expected stock returns using data of
Chinese stock market from 2000 to 2006 by applying Vassalou and Xing’s DLI to calculate
firms’ default probabilities. The empirical evidence demonstrates that the default risk is not
a systematic risk and there is no significant relationship between the expected stock returns
and the implied default risk, even controlling for size and B/M equity ratio.
Our research does not aim to find a better distress predictive model or to examine the
predictive accuracy of existing models on Chinese stock market. Since the lack of
literatures that discuss the relation between distress risk and expected stock returns in China,
we devote ourselves to explore the relationship based on the framework of Fama-French
Three-Factor Model and examine whether FF factors can proxy for distress risk calculated
through the most popular and commonly used models which have been proved able to
predict distress risk on Chinese stock market, in addition, whether an augmented four-factor
model with distress risk factor can explain expected excess stock returns better on CNAS
stock market.
3.4.1 Data
All the data needed in this study during the period July 2005 to May 2015 (119 months) is
collected from Bloomberg, and in research of both methods, we exclude the financial firms,
the capital structure of which is distinguished from that of common ones, and the firms
with negative B/P ratios also removed from the sample as Chapter 1. Since our research
focuses on the healthy firms, the firms which labeled as ‘ST’ and ‘PT’ are also out of our
43
China Securities Regulatory Commission carries out a ‘Special Treatment’ (ST) warning
mechanism to indicate abnormalities in a listed companies’ financial status Geng et al. (2015), firms
will labeled as ST if they have negative profits for two consecutive years and the equity of
shareholders is less than the registered capital, even may be delisted from the stock market if their
performance will not improve within next three years.
141
consideration. We exclude the firm if one has incomplete data at certain months. To
proceed our research and to perform the comparison, first of all, we calculate the default
probability using accounting-based model (Ohlson’s O-score) and market-based model
(Merton’s option-pricing theory) separately. These default probabilities are then used as a
basic criterion for the distinction of firms’ distress risk.
3.4.1.1 Calculate default probability by O-score and DLI
- O-score as proxy of distress risk

Due to the limitations of Altman’s Z-score, such as the strict statistical assumption of MDA,
especially, the Z-score is rather a value measuring of financial strength with little intuitive
interpretation. We choose Ohlson’s O-score which measures the probability of financial
distress.
To be included in our sample, a firm must have two-year ahead data of net profitability for
the purpose to calculate the O-score and at least 12-month data of returns. We calculate the
O-score of stocks formula (3.4) with our database at the end of each December during the
research period July 2005 to May 2015. Particularly, consistent with Dichev (1998), we do
not adjust for the ‘GNP price-level index’ of the first ‘SIZE’ variable in formula (3.4)
because the tests in this study employ monthly data, within which the index is fixed. Thus
the ‘SIZE’ in the formula (3.4) equals to log (total assets).
- DLI as proxy of distress risk

Following Vassalou and Xing (2004), we compute DLI (equation (3.8)) by applying the
methodology developed by Black and Scholes (1973) and extended by Merton (1974) to
estimate default probabilities of the individual firms on CNAS stock market. There are two
main reasons why we use Merton’s theory instead of using EDF of KMV model to
calculate distress risk in China: Firstly, as many researches emphasize that the empirical
distribution of KMV is deduced based on database of U.S. market, while there is not such a
distribution in China, it is not correct to apply EDF to estimate distress risk on Chinese
stock market; secondly, after empirical analysis of defaults, KMV has found that firms
most frequently reach the default point when the firm’s value approximately equals to
short-term (current) liabilities plus 50 percent of long-term liabilities. However, whether
the default point is the same on the Chinese stock market is still a question, it is not a
responsible way to calculate KMV’s DD based on their default point.
To calculate DLI, we need to solve Black-Scholes formula (3.8) and obtain two unknown
variables V A (asset value) and  A (asset volatility), with input value of equity VE (market
142
capitalization, equals to share price times the shares outstanding), strike price X which is
the book value of liability, risk-free interest rate r and maturity T. We follow the approach
of Moody’s KMV Crosbie and Bohn (2003) and Vassalou and Xing (2004), applying a
more complex iterative procedure to solve for the asset volatility  A and value of firm’s
asset V A :
- Step 1, at the end of each year, we calculate the daily returns of firms’ equity from
the past 12 months to obtain the volatility of equity  E , which is as the initial
estimate of  A .
- Step 2, then we use daily VE and this initial estimation  A to solve the Black-
Scholes formula on each trading day for the past year to get daily estimates V A .
- Step 3, we next take the standard deviation of those V A s as the new estimate of  A ,
which is used for the following iteration procedure.
Step 2 and 3 above are repeated until the value of  A from two consecutive iterations
converge with the tolerance level 0.001. Once we obtain the converged value of  A , we use
it to find out our final estimation V A . Then we calculate DD using equation (3.9) at each
year-end over our sample period, DLI is then denoted by the cumulative density function of
the standard normal distribution of subtractive DD (equation (3.10)).
Furthermore, we always calculate default probability with a one-year horizon. The risk-free
rate also needs to be a yearly rate, here we use one-year fixed deposit rate which is
officially determined by the People’s Bank of China. There are at least three reasons why
the time to maturity is set to one year: first, choosing one year permits comparability with
prior research such as Griffin and Lemmon (2002), Vassalou and Xing (2004) and Lin and
Chen (2008); second, a maturity of greater than one year is difficult to justify as there are
too many factors (firm specific and economy-wide) that may affect a firm’s DP over time;
third, one-year maturity represents a reasonable balance between the weight placed on
market leverage, asset volatility, and asset growth rate in the construction of the DPs
Gharghori et al. (2006).
3.4.1.2 Summary description of O-score and DLI
Our research period is from 2005 to 2014, mainly because the number of firms that have
data of O-score is rare before 2005. The annual available number of firms for which O-
score and DLI could be calculated are displayed in Table 3.2, it is obvious that there are
143
few firms which have available data of O-score before 2005 (0 for the year 2002 and 2003,
7 for the year 2004).
Table 3.2 Annual available number of firms whose O-score and DLI can be calculated
This table presents the annual number of firms which have available O-score and DLI from the year
2002 to the year 2014, “-” indicates the data which is not available.
Year O-score DLI Year O-score DLI
2002 0 - 2009 1178 1370
2003 0 - 2010 1205 1389
2004 7 822 2011 1226 1676
2005 938 1109 2012 1252 1997
2006 1013 1146 2013 1285 2193
2007 1128 1160 2014 1356 2259
2008 1151 1255
The annual aggregate O-score and aggregate DLI from 2005 to 2014 are presented in
Figure 3.2 and Figure 3.3 respectively, from which we could see that the tendency of both
figures is similar. Both the aggregate O-score and aggregate DLI increase dramatically
from 2008 to 2009, the period which is well known as the recession period because of the
worldwide financial crisis, while the aggregate default probability decreases sharply (2009-
2010) during the period of economic recovery.
Figure 3.2 Annual aggregate O-score (2005-2014)
144
Figure 3.3 Annual aggregate DLI (2005-2014)
3.4.2 Construction of distress factor and portfolios on Chinese market
3.4.2.1 Portfolios construction using O-score and DLI as proxy of distress risk
In Chapter 1, we sort stocks both into 6 (2x3) and 25(5x5) Size-B/P portfolios and we find
that the different sorts of portfolios do not affect the empirical results. In this chapter, we
construct 18 (2x3x3) Size-B/P-O-score (SBPO) portfolios as follows:
Similarly, like FF six portfolios, we sort stocks into two size portfolios (Small and Big) and
three B/P portfolios (Low, Medium and High). The size breakpoint for year t is the median
CNAS market capitalization at the end of June of year t. B/P ratio for June of year t is the
book value of equity for the last fiscal year end in t-1 divided by the price for December of
t-1, the breakpoints are the 30th and 70th CNAS percentiles. At the end of each year t-1, O-
score for year t is calculated using Ohlson’s formula (3.4) for each stock, then stocks are
sorted into 3 distress risk (DR) groups (O1, O2 and O3, which represent the low, medium
and high O-score groups, separately) according to their O-score with the breakpoint 30th
and 70th percentiles.
The 18 portfolios which are constructed at the end of each June of year t, are the
intersection of two size portfolios, three B/P portfolios, and three O-score portfolios, we
denote the portfolio which has small size, low B/P ratio and low DR as SL1, the portfolio
which has small size, low B/P ratio and medium DR as SL2, and so on the 18 portfolios are
denoted by SL1, SL2, SL3, SM1, SM2, SM3, SH1, SH2, SH3, BL1, BL2, BL3, BM1, BM2,
BM3, BH1, BH2, and BH3. Portfolios are reconstructed in June of each year. The distress
145
risk factor DRF is the difference between average returns of high O-score portfolios and
low O-score portfolios.
Following exactly the same process, we construct 18 Size-B/P-DLI (SBPD) portfolios with
two size portfolios, three B/P portfolios and three DLI portfolios (D1, D2, D3 are the low
DLI portfolio, medium DLI portfolio, and high DLI portfolio, respectively). The three DLI
portfolios are constructed in the same way that three O-score portfolios are done above,
except that the basis criterion for distinguishing firms’ distress risk is DLI instead of O-
score.
3.4.2.2 Construction of distress risk factor
The proxies of distress risk we use are the DP calculated by O-score and DLI stated in the
previous section. At the end of each June of year t, we sort stocks according to their O-
score (or DLI) of the end of year t-1 into three groups, denoted as O1, O2 and O3 (or DLI1,
DLI2, DLI3) from low to high default probabilities. The breakpoints are 30% and 70%
percentiles of the O-score (or DLI) of the sample firms. The portfolios remain
unchangeable from July of year t to June of year t+1, and portfolios are reformed at the end
of June of year t+1, and so forth for the whole sample period. The mimicking portfolio
DRF is the distress risk factor, which measures the return difference between the high-DP
portfolio (O3 or DLI3) and the low-DP portfolio.
3.5 Distress risk, size and B/P ratio of Chinese stock market
Numerous studies suggest that a firm distress risk factor could be behind the size and the
value effects. For instance, Griffin and Lemmon (2002) and Vassalou and Xing (2004) both
show that the B/M and size effects are concentrated in high default risk firms, thus leding to
the conjecture that the value and size effects are closely related to distress risk.
We aim to investigate whether there exist size and value effects on CNAS stock market
related to the distress risk in this section. In order to compare whether different methods of
measuring distress risk have different impact to our empirical results, we apply both of the
leading models, accounting-based model of Ohlson’s O-score and market-based model
according to Merton’s option-pricing theory, to estimate the distress risk of Chinese listed
firms on A-share stock market.
146
3.5.1 Distress risk and variation of stock returns
Following Griffin and Lemmon (2002) and Vassalou and Xing (2004), stocks are ranked
each June of year t according to their previous December O-score or DLI. To analyze the
relation between distress risk (DR) and stock returns, and whether there exist size and value
effects on CNAS stock market, the stocks are sorted separately into quintile and decile
according to their O-score and DLI (proxies of distress risk), we then examine whether the
portfolios with different distress risk provide significantly different returns. A significant
difference in the returns would indicate that default risk may be important for the pricing of
equities Vassalou and Xing (2004).
The simple sorts of stocks based on their O-score and DLI are presented in Panel A and
Panel B of Table 3.3, respectively. In each panel, stocks are firstly divided into five
portfolios according to their O-score or DLI, the returns, size, and B/P ratio of portfolios
are also shown in the table. Independently, we also sort stocks into 10 portfolios by their O-
score or DLI. In Panel A, no matter for the five or the ten O-score portfolios, the results are
similar. We find no obvious differences in average returns across the portfolios, and the
return difference between the highest DR portfolio and the lowest distress risk portfolio is
not statistically significant (0.0029 for the five portfolios with t-stats -1.0688 and 0.0037
for the ten portfolios with t-stats 0.9706). However, it seems that the firm size is negatively
related to the distress risk, the higher DR portfolios have smaller size. B/P ratios do not
show much differences across the DR quintile or decile except for the highest DR portfolios
which have lowest B/P ratio among other DR portfolios.
In Panel B, there is neither significant return difference among the five portfolios, nor the
ten portfolios sorted by DLI (-0.0028 for the five DLI-sorted portfolios with t-stats -0.7907
and -0.0038 for the ten DLI-sorted portfolios with t-stats -0.9025). Different from portfolios
that sorted by O-score, the trend of size-change is ruleless, but consistent with the results of
Vassalou and Xing (2004), the B/P ratio of portfolios increases for both sorts with their
default probability (proxy by DLI) increases.
Since the non-significant return differences among the portfolios sorted by both O-score
and DLI, we cannot confirm that distress risk has an impact on average stock returns.
However, the results reveal that there exist probably size or value effect associated with DR.
147
148
Table 3.3 Characteristics of portfolios sorted by O-score and DLI (2005-2014)
Panel A and Panel B of the table represent characteristics of portfolios sorted by O-score and DLI, respectively. Across the rows are the five
quintiles or ten deciles of portfolios sorted according to firm’s O-score (Panel A) or DLI (Panel B); the last column reports the return difference
between high and low default probability portfolios, numbers in parentheses are the corresponding t-statistics at 5% confidence level. Across
the rows are the portfolios’ average return, O-score (Panel A) or DLI (Panel B), size, and B/P ratio, separately. The unit of firm size is 100
million yuan, the values of DLI are reported in percentage values for easy reading.
Default High-
Low DP 2 3 4 5 6 7 8 9 High DP
Probability Low
Panel A: Characteristics of portfolios sorted by O-score
0.0029
Return 0.0228 0.0226 0.0226 0.0223 0.0258
(1.0688)
O-score -2.0799 -0.5225 0.3394 1.1449 3.8525
Size 10614 12141 7242 6296 3587
B/P 0.4293 0.4522 0.4695 0.4709 0.3540
0.0037
Return 0.0224 0.0233 0.0220 0.0232 0.0228 0.0224 0.0224 0.0223 0.0253 0.0261
(0.9706)
O-score -2.7923 -1.3728 -0.7618 -0.2833 0.1293 0.5500 0.9396 1.3501 1.8953 3.7335
Size 9515 11669 13706 10548 8203 6283 7045 5547 4675 2433
B/P 0.4139 0.4441 0.4482 0.4563 0.4700 0.4688 0.4843 0.4575 0.4128 0.2725
Panel B: Characteristics of portfolios sorted by DLI
-0.0028
Return 0.0248 0.0256 0.0234 0.0242 0.0220
(-0.7907)
DLI (%) 0.0161 0.1735 0.5499 1.4092 6.4096
Size 7950 8227 7309 7098 8012
B/P 0.3456 0.3853 0.4201 0.4560 0.5296
-0.0038
Return 0.0256 0.0240 0.0264 0.0247 0.0240 0.0227 0.0242 0.0243 0.0222 0.0219
(-0.9025)
DLI (%) 0.0021 0.0299 0.1107 0.2363 0.4207 0.6798 1.0816 1.7365 3.0395 9.7876
Size 7156 8736 9877 6579 6522 8097 7350 6841 7098 8928
B/P 0.3270 0.3641 0.3806 0.3901 0.4136 0.4266 0.4444 0.4677 0.5102 0.5490
3.5.2 Size effect
In order to examine whether there exists size effect across the distress risk groups, we
implement the two-way sorts the way Vassalou and Xing (2004) do, stocks are first sorted
into five quintiles based on their O-score or DLI (distress risk), then within each DR groups,
stocks are sorted into five size groups according to their total market capitalization.
Furthermore, we also examine the size effect across the whole sample by sorting all stocks
into five size quintiles.
The results of the 25 O-score/Size portfolios are presented in Table 3.4, across the columns
are the five DR quintiles, and across the rows are the five size quintiles. The average
portfolio returns, size, B/P ratio and O-score are displayed respectively in Panel A, B, C
and D. The results in Panel A suggest that there are significant size effects across each DR
quintile, the strongest effect is in the group of highest DR with return difference between
small and big size portfolios of 0.0194 (with t-stats 3.7498 at 5% confidence level). And we
also find the size effect across our whole sample with return difference of 0.0159 (t-stats
3.1086).
The results in Panel B show that across each size quintile, the DR of firms increases with
the size decreases apart from the lowest O-score portfolios, consistent with our findings in
Table 3.3, the smaller size firms tend to have higher default probabilities. From Panel D,
we cannot tell obvious variations within the DR quintiles; only within the highest DR
quintile, the smallest size portfolio has the highest O-score, while the biggest size portfolio
has the lowest O-score. Panel C indicates that within each size quintile, portfolios with
highest DR tend to have lower B/P ratio at least for the three smaller size quintiles, but it is
not clear that whether B/P ratio is related to DR directly from this analysis, we will carry
out the analysis of value effect subsequently.
To examine whether the size effect controlled by distress risk which estimated using
market-based method (DLI) is distinct from that controlled by O-score, we proceed the
same steps above except the stocks are first sorted into five portfolios based on their DLI
instead of O-score, then within each DLI quintiles stocks are sorted into five size groups,
the results are shown in Table 3.5. The same results are found in Panel A that across each
DR quintiles, average return of portfolio decreases as the firm size increases, and there exist
strong size effect across each DR quintile (the return differences are all statistically
significant at 5% confidence level). The strongest effect still is in the highest DR groups
(return difference is 0.0180 with t-stats 3.3040). The relationship between DR and size
seems chaotic (Panel B) except the two small size quintiles, within which the size increases
as DR increases. Panel C of Table 3.5 shows no explicit relationship between B/P ratio and
149
Table 3.3 Size effect controlled by distress risk (O-score as proxy), 2005-2014
This table presents the results of size effect controlled by distress risk proxy by O-score on Chinese
A-share stock market during 2005 to 2014. Stocks are first sorted into five O-score quintiles and
then within each O-score portfolio, stocks are sorted into five size quintiles based on their market
capitalization, thus we have got 25 O-score/Size portfolios. The average returns, size, B/P ratio and
O-score of 25 portfolios are listed in Panel A, B, C, and D respectively. Across the rows of each
panel are the five size quintiles, and across the columns of each panel are five O-score quintiles.
The unit of size is 100 million ‘yuan’.
DR Size
Small 2 3 4 Big Small-Big
Panel A: average return
0.0146
Low-O 0.0296 0.0258 0.0216 0.0197 0.0150
(2.5876)
0.0133
2 0.0284 0.0250 0.0225 0.0209 0.0151
(2.6192)
0.0136
3 0.0290 0.0246 0.0208 0.0220 0.0154
(2.6722)
0.0126
4 0.0316 0.0266 0.0182 0.0161 0.0190
(2.1924)
0.0194
High-O 0.0368 0.0296 0.0218 0.0225 0.0174
(3.7498)
Whole 0.0159
0.0327 0.0260 0.0231 0.0201 0.0169
sample (3.1086)
Panel B: size
Low-O 1399 2397 3765 6769 39072
2 1417 2415 3691 6434 46965
3 1378 2247 3433 5912 23318
4 1300 2119 3135 5312 19770
High-O 958 1484 2100 3264 10170
Panel C: B/P ratio
Low-O 0.4693 0.4961 0.4400 0.3805 0.3581
2 0.4701 0.4693 0.4586 0.4295 0.4331
3 0.4385 0.5049 0.4778 0.4685 0.4566
4 0.4339 0.4828 0.4699 0.4747 0.4921
High-O 0.2640 0.3148 0.3615 0.3835 0.4218
Panel D: O-score
Low-O -2.2624 -1.9399 -2.0330 -1.9983 -2.0299
2 -0.5313 -0.5229 -0.5187 -0.5239 -0.5201
3 0.3583 0.3522 0.3299 0.3379 0.3188
4 1.1629 1.1415 1.1554 1.1431 1.1221
High-O 3.3834 2.6381 2.6904 2.2988 2.0960
150
Table 3.4 Size effect controlled by distress risk (DLI as proxy), 2005-2014
This table presents the results of size effect controlled by distress risk proxy by DLI on Chinese A-
share stock market during 2005 to 2014. Stocks are first sorted into five DLI quintiles and then
within each DLI quintile, stocks are sorted into five size quintiles based on their market
capitalization, thus we have got 25 DLI/Size portfolios. The average returns, size, B/P ratio and DLI
of 25 portfolios are listed in Panel A, B, C, and D respectively. Across the rows of each panel are
the five size quintiles, and across the columns of each panel are five DLI quintiles. The values of
DLI are percentage values, and the unit of size is 100 million ‘yuan’.
DR Size
Small 2 3 4 Big Small-Big
0.0163
Low-DLI 0.0323 0.0289 0.0250 0.0209 0.0160
(3.1322)
0.0116
2 0.0323 0.0259 0.0248 0.0229 0.0207
(2.1045)
0.0127
3 0.0289 0.0270 0.0243 0.0196 0.0162
(2.3553)
0.0160
4 0.0328 0.0260 0.0259 0.0182 0.0167
(3.1280)
0.0180
High-DLI 0.0339 0.0220 0.0194 0.0183 0.0159
(3.3040)
Panel B: size
Low-DLI 1160 1919 3024 5102 28769
2 1219 1922 2819 4587 30812
3 1223 1905 2797 4718 26083
4 1259 2013 2947 5024 24415
High-DLI 1477 2412 3714 6357 26289
Panel C: B/P ratio
Low-DLI 0.3775 0.3898 0.3675 0.3129 0.2799
2 0.4017 0.4126 0.3923 0.3720 0.3477
3 0.3926 0.4443 0.4386 0.4217 0.4033
4 0.4192 0.4715 0.4768 0.4578 0.4548
High-DLI 0.4710 0.5367 0.5409 0.5371 0.5623
Panel D: DLI (%)
Low-DLI 0.0153 0.0205 0.0178 0.0124 0.0142
2 0.1712 0.1674 0.1843 0.1716 0.1730
3 0.5558 0.5494 0.5640 0.5399 0.5411
4 1.3873 1.3848 1.4036 1.4245 1.4464
High-DLI 5.4074 5.9529 6.3253 6.4563 7.9216
151
distress risk, but within each size quintile, the portfolio with highest DR seems has the
higher B/P ratio. The most obvious variation is across the highest DR group in panel D, the
highest distress risk belongs to the big size portfolio (Big-5). Overall, average returns are
closely related to firm size in both Table 3.4 and Table 3.5. it appears that there exists
strong size effect across DR group and in the whole sample as well, no matter the proxy for
DR is O-score or DLI. This finding is consistent with our empirical results in Chapter 1.
However, the relation between size (or B/P ratio) and DR is kind of chaotic and not easy to
distinguish from the size effect tests controlled by DR.
3.5.3 Value effect
Similar to the analysis of size effect, to examine the value effect across the DR groups,
stocks are first sorted into five DR quintiles and within each DR quintile, the stocks are
sorted into five B/P groups. Across the rows of Table 3.6 are the five B/P quintiles and
across the columns are the five O-score quintiles. In Panel A, we find the return difference
between the highest B/P quintile and the lowest B/P quintile neither significant for all of the
five DR portfolios nor significant across the whole sample, which indicates that there exists
probability no value effect controlled by distress risk on CNAS stock market during the
research period.
In Panel B, the relationship between B/P ratio and DR is not clear. The DR is negatively
related to firm size within the lowest B/P quintile (Panel C). In Panel D, across each DR
quintile, the relationship between O-score and B/P ratio is not clear except for the lowest
and highest DR portfolios. Across the lowest DR quintile, it seems that with tiny change of
O-scores, higher B/P ratio associated with higher O-score; but across the quintile which has
the highest DR, the higher the B/P ratio the lower O-score of portfolios, exceptionally, the
portfolio with lowest B/P ratio has the highest DR (with O-score 3.1856).
The results of value effect controlled by distress risk which is proxy by DLI are displayed
in Table 3.7. Notice that in Panel A, same as Table 3.6, none of the return difference across
the five DLI quintile is significant. In Panel B, within each B/P quintile, higher DLI (DR)
portfolios tend to have higher B/P ratio. In panel D, the DLI of portfolio increases as the
B/P ratio increase in the highest DR quintile except for the lowest B/P portfolio.
152
Table 3.5 Value effect controlled by distress risk (O-score as proxy)
This table presents the results of value effect controlled by distress risk proxy by O-score on
Chinese A-share stock market during 2005 to 2014. Stocks are first sorted into five O-score
quintiles and then within each O-score quintile, stocks are sorted into five B/P quintiles based on
their book-to-price ratio, thus we have got 25 O-score/B/P portfolios. The average returns, B/P ratio
size, and O-score of 25 portfolios are listed in Panel A, B, C, and D respectively. Across the rows of
each panel are the five B/P ratio quintiles, and across the columns of each panel are five O-score
quintiles. The unit of the size is 100 million yuan.
DR B/P ratio
Low-B/P 2 3 4 High-B/P High-Low
-0.0013
Low-O 0.0230 0.0206 0.0237 0.0243 0.0217
(-0.2722)
0.0052
2 0.0177 0.0246 0.0245 0.0228 0.0229
(1.4291)
0.0026
3 0.0200 0.0253 0.0234 0.0215 0.0226
(0.7053)
0.0026
4 0.0201 0.0205 0.0232 0.0238 0.0227
(0.5363)
0.0037
High-O 0.0216 0.0253 0.0249 0.0278 0.0253
(0.8270)
Whole 0.0017
0.0220 0.0248 0.0245 0.0253 0.0237
sample (0.4963)
Panel B: B/P ratio
Low-O 0.1723 0.2929 0.3923 0.5109 0.7808
2 0.1921 0.3159 0.4150 0.5390 0.8007
3 0.1879 0.3252 0.4322 0.5632 0.8399
4 0.1901 0.3221 0.4245 0.5551 0.8649
High-O 0.0809 0.2076 0.3203 0.4471 0.7178
Panel C: size
Low-O 14998 8727 6297 10278 11381
2 8873 12183 9712 15495 14865
3 8497 5994 7845 6364 7379
4 6116 5952 5638 5787 7842
High-O 2565 3507 3981 4125 5225
Panel D: O-score
Low-O -2.3082 -2.0629 -2.0357 -1.9592 -1.9006
2 -0.5188 -0.5231 -0.5225 -0.5127 -0.5454
3 0.3479 0.3297 0.3343 0.3387 0.3445
4 1.1593 1.1657 1.1459 1.1256 1.1352
High-O 3.1856 2.4067 2.2171 2.1017 2.0868
153
Table 3.6 Value effect controlled by distress risk (DLI as proxy)
This table presents the results of size effect controlled by distress risk proxy by DLI on Chinese A-
share stock market during 2005 to 2014. Stocks are first sorted into five DLI quintiles and then
within each DLI quintile, stocks are sorted into five size quintiles based on their market
capitalization, thus we have got 25 DLI/Size portfolios. The average returns, size, B/P ratio and DLI
of 25 portfolios are listed in Panel A, B, C, and D respectively. Across the rows of each panel are
the five size quintiles, and across the columns of each panel are five DLI quintiles. The value of
DLI is percentage value and the unit of the size is 100 million yuan.
B/P ratio
DLI Low-B/P 2 3 4 High-B/P High-Low
0.0043
Low-DLI 0.0218 0.0234 0.0256 0.0265 0.0261
(1.1785)
0.0032
2 0.0219 0.0242 0.0262 0.0297 0.0251
(0.9650)
0.0052
3 0.0196 0.0257 0.0224 0.0237 0.0249
(1.6248)
0.0030
4 0.0220 0.0228 0.0241 0.0267 0.0250
(0.9643)
-0.0018
High-DLI 0.0243 0.0222 0.0206 0.0203 0.0225
(-0.4368)
Panel B: B/P ratio
Low-DLI 0.1317 0.2355 0.3256 0.4178 0.6188
2 0.1610 0.2736 0.3570 0.4559 0.6812
3 0.1769 0.2999 0.3909 0.4984 0.7365
4 0.1927 0.3306 0.4278 0.5427 0.7887
High-DLI 0.2249 0.3736 0.4893 0.6337 0.9287
Panel C: size
Low-DLI 11341 7380 5650 9375 5955
2 6785 5533 7639 6090 15186
3 7068 7924 6777 6988 7762
4 6366 7909 6760 5386 9096
High-DLI 7459 7905 7956 7192 9554
Panel D: DLI (%)
Low-DLI 0.0105 0.0127 0.0190 0.0174 0.0206
2 0.1682 0.1721 0.1765 0.1777 0.1731
3 0.5299 0.5535 0.5551 0.5553 0.5557
4 1.3922 1.3736 1.3849 1.4307 1.4652
High-DLI 6.2900 5.7332 6.5571 6.6960 6.7771
154
We summarize that there exists strong size effect but no value effect, however, the robust
size effect and lack of value effect are robust when controlled by distress risk as well as for
the whole sample on CNAS stock market from 2005 to 2014. And the effects are also
robust to the different proxies for distress risk. Nevertheless, the relationship between size
and DR, and B/P ratio and DR are quite confusing. The most distinct is that B/P ratio is
positively related to DR that use DLI as proxy (Panel B of Table 3.7), portfolios with
higher DLI have higher B/P ratio.
3.6 Augmented four-factor model and empirical evidence
In order to investigate whether FF three factors are proxies for distress risk, we examine an
augmented four-factor model, which include market factor (market premium), size factor
(SMB), value factor (HML) and distress risk factor (DRF). If all the priced information in
SMB and HML is related to financial distress risk, we would expect to find that in the
presence of DRF, SMB and HML lose all their ability to explain equity returns 44 . In
particular, we use both accounting-based model and market-based model as proxy for
distress risk to investigate whether different methods of estimating default probability cause
different results.
3.6.1 Augmented four-factor model
We add a distress risk factor on the basis of FF3F model to form the augmented four-factor
model:
Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  di DRF  i ,t (3.18)
where DRF is our mimicking portfolio, distress risk factor, di is the regression coefficient
of DRF, and i ,t is the error term for portfolio i at time t.
DRF is constructed separately using O-score and DLI as proxy for distress risk, denoted as
DRFO score and DRFDLI , the summary statistics of four factors are shown in Table 3.8 Panel
A presents the statistic description of four factors, the time-series mean of market premium
(0.0024), SMB (0.0119) and DRFO score (0.0010) are positive, while that of HML (-0.0020)
44
Vassalou and Xing (2004)
155
Table 3.7 Summary statistics of four factors: market factor, SMB, HML and distress risk
factor (July 2005 to May 2015)
This table presents summary statistics of four factors, market factor, size factor, value factor and
distress risk factor ( DRFO score or DRFDLI ). Statistic description of these factors are shown in Panel
A and the correlation matrix among these factors is presented in Panel B.
RM  R f SMB HML DRFO score DRFDLI
Panel A Statistic description of four factors
Mean 0.0024 0.0119 -0.0020 0.0010 -0.0026
Standard Error 0.0083 0.0038 0.0030 0.0019 0.0030
t-statistic 0.2892 3.1316 -0.6667 0.5263 -0.8667
Median 0.0064 0.0137 -0.0057 0.0011 -0.0014
S.D. 0.0905 0.0414 0.0329 0.0206 0.0326
Sample Variance 0.0082 0.0017 0.0011 0.0004 0.0011
Kurtosis 1.1923 2.3341 4.0778 -0.2535 2.0064
Skewness -0.7299 -0.7784 0.5098 0.2176 0.1691
Panel B Correlation matrix of four factors
RM-RF 1
SMB -0.0175 1
HML 0.1985 -0.3246 1
DRFO score 0.2362 0.2806 0.3672 1
DRFDLI 0.4040 -0.2983 0.7139 0.6665 1
and DRFDLI (-0.0026) are negative. The only significant t-stats is from SMB, and the size
factor is most likely to be priced in equity returns.
The correlation matrix of the four factors in Panel B indicate that the two DRF factors are
highly correlated (with correlation coefficient 0.6665), which means that DRFO score and
DRFDLI may contain most of the same information. It is worth noting that DRFDLI and
HML are highly positive correlated, the component of equity returns that each factor
explains may be similar.
3.6.2 Time-series regressions analysis
We carry out the TSRs of the augmented four-factor model on CNAS stock market in this
section. To construct distress risk factor, we apply both accounting-based model and
market-based model to estimating default probability of stocks.
156
3.6.2.1 Accounting-based distress risk factor and regression results
Table 3.8 shows the summary statistics of 18 portfolios sorted on Size, B/P and O-score.
For all the 18 portfolios, the average excess returns are lower for small size portfolios while
higher for big size portfolios, which is consistent with the size effect conclusion that equity
return is negatively related to firm size, the big size portfolios tend to have lower average
returns. However, we find no obvious relationship between average excess returns and B/P
ratio or distress risk proxy by O-score, except the highest B/P portfolios with small size
(across the highest B/P group, the higher DR portfolio tends to have higher average excess
return) and the lowest DR portfolios with small size (across the lowest DR portfolios,
higher B/P portfolio tends to have lower average excess return, which is inconsistent with
FF’s findings that average stock returns are positively related to B/M ratio). The results of
Table 3.9 also reveal that distress risk is related to firm size, regarding to the size parts, the
higher DR portfolios always have smaller size (except the portfolio SH1), meanwhile,
regarding to the O-score parts, the big size portfolios have lower O-score than the small
size portfolios excluding the lowest DR groups. Whereas, it is not easy to tell the
relationship between B/P ratio and O-score from the analysis so far.
To investigate how the four factors, explain average excess returns of 18 SBPO sorted
portfolios, we following perform the TSRs and the results are presented in Table 3.9. The
left-hand part presents the regression coefficients and adjusted R-squares, while the right-
hand part presents the corresponding t-statistics corrected for heteroscedasticity and
autocorrelation using Newey-West estimator with five lags and the residual standard error.
As always, the coefficients of excess market return and SMB are all highly significant at 5%
confidence level, and there exists size effect. Regression coefficients β(h) increase as B/P
ratio increase, but the loadings on HML seem only to be significant within the highest B/P
groups (for small size portfolios and big size portfolios) and the lowest B/P group (for big
size portfolios). Loadings on DRF β(d) increase as O-score increase except for the portfolio
group with small size and medium B/P ratio, but the significant coefficients mainly exist in
the lowest (for big size portfolios) and highest (for small size and big size portfolios) O-
score groups. Comparing with the regressions on FF3F model of the same 18 portfolios
(Appendix F), the TSR coefficients of FF three factors do not have a noteworthy difference
with or without an augmented factor DRF in addition to FF3F model. Adding a DRF factor
(with averaged adjusted R-square 0.8990) does not change markedly the time-series
explanatory power of FF three factors (with averaged adjusted R-square 0.8939).
157
Table 3.8 Summary statistics of 18 (2x3x3) Size-B/P-O-score sorted portfolios (July 2005-May
2015)
The 18 portfolios are constructed from the intersection of three independent sorts of all stocks into
two size, three B/P and three DR portfolios. DR is estimated by O-score. The table has two parts:
small size and big size, and in each part, average excess returns, firm numbers, size, B/P ratio and
O-score of 18 portfolios are presented. Across the rows are the three B/P portfolios (L, M and H),
and across the columns are the three O-score portfolios (O1, O2 and O3). The unit of size is 100
million ‘yuan’.
O-score B/P ratio
L M H L M H
Small size
Average excess returns Firm numbers
O1 0.0190 0.0170 0.0154 24 55 49
O2 0.0129 0.0165 0.0161 41 76 73
O3 0.0153 0.0174 0.0173 79 63 46
Size B/P ratio
O1 1874 1875 1887 0.2161 0.3970 0.6941
O2 1804 1849 1939 0.2090 0.3955 0.6958
O3 1619 1800 1889 0.1675 0.3869 0.6824
O-score
O1 -1.8157 -1.5943 -1.5123
O2 0.3383 0.3118 0.3212
O3 2.5651 1.8517 1.7261
Big size

O1 0.0078 0.0086 0.0066 72 74 56
O2 0.0090 0.0102 0.0059 69 89 94
O3 0.0072 0.0063 0.0095 46 51 45
Size B/P ratio
O1 14595 14823 25960 0.2035 0.3924 0.7028
O2 11436 11108 13714 0.2112 0.3977 0.7284
O3 7681 8297 8762 0.1877 0.3957 0.7198
O-score
O1 -1.7393 -1.5941 -1.5076
O2 0.2410 0.2565 0.2837
O3 1.9355 1.6339 1.6183
158
Table 3.9 Time-Series Regressions of 18 Size-B/P-O-score sorted portfolios on augmented

four-factor model (July 2005-May 2015)
This table reports the time-series regression results for 18 portfolios formed on size, B/P ratio and
O-score. The stocks are divided into two size groups based on the breakpoint of median market
capitalization, and the breakpoints for the three B/P groups are the top 30%, median 40% and
bottom 30% of B/P ratio, similarly as B/P groups, the stocks are divided into three groups based on
their O-score (O1, O2 and O3, represent the low, medium and high O-score separately). The
intersection of two size groups, three B/P groups and O-score groups form 18 portfolios. The left
part of the table is the coefficients of time-series regressions and adjusted R-square, the right part is
the corresponding t-stats corrected for heteroscedasticity and autocorrelation using Newey-West
estimator and residual standard error, and the numbers in bold are significant at 5% confidence level.
Regression: Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  d i DRFO-score  i ,t
Small size
O-score B/P ratio
L M H L M H
a t (a )
O1 0.0000 -0.0016 -0.0043 0.0003 -0.6528 -1.5354
O2 -0.0070 -0.0020 -0.0018 -1.9398 -0.9184 -0.6258
O3 -0.0047 -0.0003 0.0006 -1.7372 -0.1185 0.2581
b t (b)
O1 0.9835 1.0019 0.9784 18.1800 20.2270 21.1813
O2 1.0186 0.9969 1.0041 19.9062 23.4918 17.6368
O3 0.9647 0.9710 0.9754 22.3848 22.0340 18.9126
s t (s)
O1 1.4076 1.3903 1.5692 10.6830 11.8585 13.3027
O2 1.4445 1.3948 1.3712 11.2147 13.6715 10.6478
O3 1.3875 1.2284 1.1975 11.4363 11.7114 10.4843
h t (h)
O1 -0.1103 0.1326 0.5809 -0.6975 0.9491 3.6733
O2 -0.0190 0.2152 0.4466 -0.1134 1.6752 2.8900
O3 -0.0932 0.0641 0.3075 -0.6026 0.5211 2.1046
d t (d)
O1 -0.2854 -0.0707 -0.1270 -1.4096 -0.4613 -0.8243
O2 0.2218 -0.1139 0.0635 1.6652 -0.7966 0.3578
O3 0.8797 0.8008 0.6211 6.7380 7.0318 3.4913
Adj R-square Residual standard error
O1 0.8684 0.9114 0.9033 0.0403 0.0330 0.0356
O2 0.9002 0.9175 0.9024 0.0368 0.0316 0.0354
O3 0.9233 0.9199 0.9067 0.0320 0.0318 0.0342
159
Table 3.9 Continued
Big size
O-score B/P ratio
L M H L M H
a t (a )
O1 -0.0022 -0.0034 -0.0047 -0.9766 -1.2719 -1.8308
O2 -0.0038 -0.0004 -0.0045 -1.2982 -0.1281 -1.7928
O3 -0.0054 -0.0060 -0.0001 -1.7151 -2.0373 -0.0473
b t (b)
O1 0.9715 1.0489 1.0288 18.0644 19.9291 18.9004
O2 1.0192 1.0881 1.0399 18.1832 20.2544 17.9173
O3 1.0165 1.1039 1.0602 16.4040 17.4080 24.4310
s t (s)
O1 0.6021 0.8563 0.8358 5.3909 6.8872 8.3814
O2 0.8453 0.6659 0.7390 7.1385 5.1131 6.2367
O3 0.7488 0.7823 0.6491 5.7002 6.0745 5.8866
h t (h)
O1 -0.4976 0.0938 0.5270 -3.4900 0.6492 4.2189
O2 -0.2723 0.0177 0.5810 -1.7387 0.1078 3.7957
O3 -0.4012 0.0356 0.6480 -2.5041 0.2226 4.9922
d t (d)
O1 -0.4896 -0.4684 -0.0989 -3.9778 -3.5396 -0.8279
O2 -0.2171 0.0833 0.2654 -1.4397 0.5639 2.2562
O3 0.4937 0.3375 0.5305 3.0586 1.9698 3.1082
O1 0.8744 0.8848 0.9061 0.0336 0.0354 0.0324
O2 0.8752 0.8928 0.9090 0.0364 0.0354 0.0329
O3 0.8592 0.8796 0.9166 0.0402 0.0395 0.0326
Thus we conclude that there exists distress risk effect, since portfolios with higher O-score
associated with higher return. The DRF constructed using O-score as proxy for DR has
explanatory power in time-series stock returns only for the portfolios with extreme DR
(lowest DR or highest DR). Whereas, the augmented DRF add limited explanatory power
in explaining stock returns, and FF factors are not proxy for distress risk in explaining time-
series variations of average stock returns, at least when the DRF is constructed using O-
score to estimate financial distress risk. We then perform the same TSRs using DLI as
proxy for DR.
160
3.6.2.2 Market-based distress risk factor and regression results
In order to test whether different methods for predicting distress risk affect the empirical
results, we proceed the Merton’s option-pricing model (market-based model) in predicting
default probability, which is later applied by Vassalou and Xing (2004). Vassalou and Xing
denote Merton’s default probability as DLI (Default Likelihood Indicator), follow which
we calculate the default probability of CNAS stocks.
Table 3.10 displays the summary statistics of 18 SBPD sorted portfolios, the structure of
the table is similar to Table 3.8 except the proxy for distress risk is DLI instead of O-score.
Not surprisingly that average excess returns are negatively related to firm size, however,
the relationship between average excess return and B/P ratio or DLI is chaotic. The
relationship between size and DLI is not as clear as that between size and O-score, while it
seems that B/P ratio increases as DLI increases except the portfolio SL1. Next, we will
implement the TSRs to investigate how the four factors explain average excess returns and
to figure out the relationship among stock returns, size, B/P ratio and distress risk.
The TSRs’ results on the augmented four-factor model of 18 SBPD sorted portfolios are
reported in Table 3.11. It shows pretty much the same results as that of 18 SBPO sorted
portfolios in Table 3.9. All the regression coefficients for excess market return and SMB
are highly significant, and excess stock returns are negatively related to firm size.
Exceptionally, β(d), the coefficients of DRFDLI , are significant for both the lowest DLI
portfolios (with negative loadings) and the highest DLI portfolios (with positive loadings),
consistent with Vassalou and Xing (2004), we find positive relationship between excess
stock returns and distress risk on Chinese market, but only in the lowest or (and) highest
DR (DLI) portfolios. Our results are contrary to some literature45 that conclude the high
default risk is not rewarded by higher returns. Comparing with the TSRs on FF3F model of
the same 18 SBPD portfolios (Appendix F), in the presence of augmented DRF (DLI as
proxy for DR), FF factors do not lose their explanatory power, however, the four-factor
model indeed capture slightly more variation of the 18 SBPD sorted portfolios’ return (with
average adjusted R-square 0.9027) than the original FF3F Model (with average adjusted R-
square 0.8980).
We can conclude that the excess market returns and size factor SMB are always highly
positively related to excess average stock returns, and big size stocks always have lower
average returns. The value factor HML also has explanatory power in explaining the time-
series average stock returns, but only exist in the extreme (lowest or highest) B/P groups. In
addition, there exists significant distress risk effect on CNAS stock market over the sample
45
Dichev (1998), Griffin and Lemmon (2002) and Lin and Chen (2008), etc.
161
Table 3.10 Summary statistics of 18 (2x3x3) Size-B/P-DLI sorted portfolios
The 18 portfolios are constructed from the intersection of three independent sorts of all stocks into
two size, three B/P and three DR (proxy by DLI) portfolios. DR in this table is proxy by DLI. The
table has two parts: small size and big size, and in each part, average excess returns, firm numbers,
size, B/P ratio and DLI of 18 portfolios are presented. Across the rows are the three B/P portfolios
(L, M and H), and across the columns are the three DLI portfolios (D1, D2 and D3). The DLI is
percentage value, and the unit of size is 100 million ‘yuan’.
DLI B/P ratio

L M H L M H
Small size
D1 0.0176 0.0201 0.0176 70 104 52
D2 0.0135 0.0173 0.0188 80 136 94
D3 0.0206 0.0189 0.0160 37 73 82
Size B/P ratio
D1 1797 1735 1712 0.2016 0.3924 0.6429
D2 1748 1815 1838 0.1992 0.3934 0.6573
D3 1767 1871 1956 0.1955 0.3944 0.7034
DLI (%)
D1 0.0432 0.0510 0.0612
D2 0.5721 0.6073 0.6332
D3 4.4861 3.8763 4.5588
Big size
D1 0.0110 0.0096 0.0072 118 74 32
D2 0.0075 0.0102 0.0090 98 118 74
D3 0.0087 0.0064 0.0065 45 91 120
Size B/P ratio
D1 12069 15431 33975 0.1951 0.3856 0.6680
D2 10399 13418 14494 0.2113 0.3930 0.6913
D3 11296 11844 11611 0.2185 0.4021 0.7384
DLI (%)
D1 0.0419 0.0454 0.0615
D2 0.5209 0.6115 0.6679
D3 4.5184 4.9209 5.7250
162
Table 3.11 Time-Series Regressions of 18 Size-B/P-DLI sorted portfolios on augmented four

factor model (July 2005-May 2015)
In this table, the time-series regressions’ results for 18 portfolios formed on size, B/P ratio and DLI
are presented. The stocks are divided into two size groups based on the breakpoint of median
market capitalization, and the breakpoints for the three B/P groups are the top 30%, median 40%
and bottom 30% of B/P ratio (named Low, Medium and High), similarly as B/P groups, the stocks
are divided into three groups based on their DLI (named D1, D2, and D3, represent the low,
medium and high DLI separately). The intersection of two size groups, three B/P groups and DLI
groups form 18 portfolios. The left part of the table is the coefficients of time-series regressions and
adjusted R-square, the right part is the relative t-statistics corrected for heteroscedasticity and
autocorrelation using Newey-West estimator and residual standard error, and the numbers in bold
are significant at 5% confidence level.
Regression: Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  di DRFDLI  i ,t
Small size
DLI B/P ratio
L M H L M H
a t (a )
D1 -0.0029 -0.0005 -0.0039 -1.0573 -0.1862 -1.1439
D2 -0.0077 -0.0020 -0.0007 -2.5693 -0.9118 -0.2556
D3 0.0024 0.0011 -0.0007 0.7513 0.4257 -0.2726
b t (b)
D1 1.0080 1.0205 1.0038 20.9669 22.5534 22.0400
D2 0.9851 1.0132 1.0201 21.8739 23.4259 20.7465
D3 0.9412 0.9777 0.9626 20.7408 29.0933 18.9405
s t (s)
D1 1.4325 1.4235 1.5812 17.0791 13.0447 13.8918
D2 1.6249 1.4059 1.4706 14.8188 16.0954 12.1920
D3 1.4806 1.4054 1.3504 14.7955 15.2163 12.7115
h t (h)
D1 -0.0995 0.1619 0.7386 -0.5584 0.9116 3.7294
D2 -0.1802 0.0882 0.3894 -1.0067 0.6316 2.1180
D3 -0.2696 -0.0647 0.1716 -1.5502 -0.4668 0.9373
d t (d)
D1 -0.2688 -0.4577 -0.5041 -2.1132 -3.1580 -3.1573
D2 0.2716 -0.0927 -0.1000 2.0505 -0.8700 -0.6824
D3 0.6708 0.4422 0.4337 5.2632 4.0964 3.4228
D1 0.9191 0.9138 0.9065 0.0317 0.0327 0.0350
D2 0.9228 0.9283 0.9082 0.0325 0.0298 0.0348
D3 0.8985 0.9265 0.9147 0.0365 0.0304 0.0325
163
Big size
DLI B/P ratio
L M H L M H
a t (a )
D1 -0.0010 -0.0039 -0.0053 -0.4195 -1.6124 -1.7204
D2 -0.0052 -0.0023 -0.0022 -1.8547 -0.8540 -0.8066
D3 -0.0021 -0.0044 -0.0032 -0.6841 -1.5073 -1.5033
b t (b)
D1 0.9920 1.0514 0.9435 19.6819 24.5412 16.7052
D2 1.0288 1.1002 1.0605 16.2234 21.3147 23.2151
D3 1.0576 1.0880 1.0344 24.3647 20.7160 21.7020
s t (s)
D1 0.5848 0.8023 0.9166 5.6470 7.1406 8.0476
D2 0.8035 0.7937 0.7936 6.4233 7.7667 8.6943
D3 0.7138 0.7723 0.8138 7.3415 5.8087 8.7760
h t (h)
D1 -0.3030 0.0443 0.5003 -2.0710 0.2813 3.6915
D2 -0.5414 -0.0845 0.3792 -3.1207 -0.4081 2.2044
D3 -0.4844 -0.1790 0.4319 -2.9118 -0.9566 2.8856
d t (d)
D1 -0.6446 -0.4718 -0.0774 -5.1570 -3.7512 -0.6858
D2 0.1191 -0.0783 0.0341 0.8120 -0.4959 0.2478
D3 0.3748 0.4312 0.5130 2.5523 2.6643 4.7165
D1 0.8745 0.8879 0.8678 0.0338 0.0344 0.0367
D2 0.8719 0.8986 0.9098 0.0379 0.0347 0.0324
D3 0.8822 0.8883 0.9287 0.0373 0.0379 0.0300
period, the explanatory power of DRF exists in the extreme top (highest DR) and bottom
(lowest DR) portfolio groups. FF factors cannot proxy for DRF, instead, DRF explains
average stock returns in combination with FF factors on CNAS stock market. However, the
additional explanatory power of DRF is limited. Besides, the distress risk factor constructed
based on DLI seems performs better than that constructed based on O-score.
164
3.6.3 Cross-sectional regressions of the augmented four-factor model
Furthermore, we also conduct the CSRs using FM two-stage approach to examine whether
the augmented four-factor model is able of capturing the variation of the cross-sectional
average excess stock returns. The CSR formula is as follows:
Ri ,t  R f  i   M bî   SMB ˆsi   HML hî   DRF dˆ i  i ,t (3.19)
where, Ri ,t  R f is the excess returns of the same 18 Size-B/P-DR sorted portfolios in the
TSRs; b̂i , ŝi , ĥi and d̂i are the estimated coefficients of the market factor, size factor SMB,
value factor HML and distress risk factor DRF obtained from TSRs;  M ,  SMB ,  HML and
 DRF are the coefficients of CSRs.
Following equation (3.19), the CSRs are performed each month (119 months, July 2005 to
May 2015) by regressing the excess returns of 18 Size-B/P-DR sorted portfolios on the
coefficients obtained from the TSRs, thus we have 119 sets of CSR coefficients. The
gammas are then calculated by taking the simple average of the 119 sets of CSR
coefficients. Similarly, we also report both the FM t-stats and the EIV adjusted t-stats as in
previous chapters. And we also examine the four-factor model with DRF estimated based
on O-score and DLI, respectively.
In order to examine whether FF factors proxy for DRF in capturing the cross-sectional
variation of average stock returns, we perform three kinds of regressions on (1) the
augmented four-factor model as presented in equation (3.19); (2) the original FF3F Model;
and (3) only excess market return and DRF using 18 Size-B/P-DR (O-score or DLI)
portfolios.
The empirical results are reported in Table 3.12, in Panel A, the DR is estimated by O-score,
while in Panel B, the DR is measured by DLI. Each panel contains three parts, which report
respectively the CSR results of the three kinds of regressions (denoted as R1, R2 and R3)
mentioned above. The first row in each part is the average of 119 CSR coefficients, and the
corresponding FM t-stats and EIV adjusted t-stats (SH t-stats) are reported in the bracket
below the coefficients. The averaged adjusted R-squares are reported in the last columns in
percentage values.
The results in both Panel A and Panel B are much the same. No matter regressing the 18
portfolios on only market factor (R3) and DRF (both constructed based on O-score and DLI)
or on FF3F Model (R2), loadings on DRF is never an important determinant of average
returns. Furthermore, comparing the first two regressions (R1 and R2), FF factors (market
165
Table 3.12 Cross-sectional regressions of 18 Size-B/P-DR sorted portfolios on augmented four-

factor model, on FF3F Model, and on market factor and DRF (July 2005-May 2015, 119
months)
The cross-sectional regression results of 18 Size-B/P-DR sorted portfolios on the augmented four-
factor model (R1), on FF3F Model (R2), and on market factor and DRF (R3) are reported in this
table. In Panel A the DRF is constructed by O-score sorted portfolios and in Panel B the DRF is
constructed by DLI sorted portfolios. Each panel contains three parts (R1, R2, and R3), the first row
of each part reports the cross-sectional regressions’ intercepts and coefficients (gammas), the
second and third rows report the corresponding Fama-MacBeth t-statistics (FM t-stats) and Shanken
corrected t-statistics (SH t-stats). The numbers in bold are the t-statistics that are significant at 5%
confidence level. The averaged adjusted R-squared are reported in the last column in percentage
form.
Regressions1 (R1): R  R     bˆ   ˆs   hˆ   dˆ  
i ,t f i M i SMB i HML i DRF i i ,t
Regressions2 (R2): Ri ,t  R f  i   M bî   SMB ˆsi   HML hî  i ,t

Regressions3 (R3): Ri ,t  R f  i   M bî   DRF dˆ i  i ,t
Const.(α) M  SMB  HML  DRF Adj. R 2 (%)

Panel A: Cross-sectional regressions on FF3F Model and on DRF (O-score)
R1 0.0230 -0.0214 0.0101 -0.0002 0.0009 48.92
FM t-stats (1.8014) (-1.4530) (2.4481) (-0.0717) (0.4641)
SH t-stats (-2.4014) (2.5008) -0.0711 0.4622
R2 0.0255 -0.0236 0.0098 -0.0003 44.05

FM t-stats (1.9885) (-1.6110) (2.3688) (-0.1059)
SH t-stats (-2.6479) (2.4250) (-0.1042)
R3 0.0627 -0.0545 0.0032 35.62

FM t-stats (2.6825) (-2.4364) (1.4291)
SH t-stats (-5.6066) (1.2769)
Panel B: Cross-sectional regressions on FF3F Model and on DRF (DLI)
R1 0.0058 -0.0065 0.0127 -0.0025 -0.0026
FM t-stats (0.4761) (-0.4702) (3.0848) (-0.7622) (-0.8765) 57.24
SH t-stats -0.7325 (3.1387) -0.7578 -0.8735
R2 0.0038 -0.0046 0.0127 -0.0023

FM t-stats (0.2963) (-0.3298) (3.0897) (-0.7190) 50.39
SH t-stats (-0.5220) (3.1698) (-0.7087)
R3 0.0409 -0.0278 -0.0049

FM t-stats (2.4606) (-1.7412) (-1.5825) 23.07
SH t-stats (-2.9414) (-1.5804)
166
beta and SMB of Panel A, SMB of Panel B) do not lose their explanatory power in the
presence of DRF. We might conclude that though DRF is a priced factor in determining
time-series average returns, it is not the case in determining cross-sectional average returns
on CNAS stock market over the sample period.
The results also indicate that no matter the DRF is constructed using O-score or DLI as
proxy, neither value factor HML and distress risk factor DRF is able to capture the cross-
sectional variation in average stock returns. The evidence again proves the existence of size
premium on CNAS stock market.
3.7 Conclusions
We use both accounting-based model (O-score) and market-based model (DLI) to estimate
distress risk and construct a mimicking distress risk factor on CNAS stock market. We first
examine whether there exist size effect and value effect controlled by default risk or in the
whole sample during our research period in China. The results suggest that there only exists
size effect but this effect seems not related to distress risk, and this effect do not depend on
the method we use to estimate the distress risk.
Following literature such as Vassalou and Xing that investigate whether the Fama-French
factors can be proxy by distress risk factor, we augment FF3F model by adding a
mimicking distress risk portfolio (DRF). We examine the variation of excess average
returns of 18 Size-B/P-DR sorted portfolios explained by the augmented four-factor model
both through time-series and CSRs. From the TSRs, we conclude that the explanatory
power of FF three factors do not has significant change with or without the DRF is
presented in the model. FF factors cannot proxy for DRF, instead, DRF explains time-series
average stock returns combining with FF factors on CNAS stock market; most
contributions to the excess average returns of value factor HML and distress risk factor
DRF are concentrated in the extreme top or bottom portfolios, and the average returns are
positively related to the distress risk. However, the additional explanatory power of DRF is
limited. Comparing the regression results by using O-score and DLI as proxy of distress
risk, the distress risk factor constructed based on DLI seems performs slightly better than
that constructed based on O-score.
We provide evidence from the CSRs that no matter the DRF is constructed using O-score
or DLI as proxy, there exists robust size premium (robust market premium when DR is
167
estimated using O-score), neither value factor HML and distress risk factor DRF is able to
capture the cross-sectional variation in average stock returns. Thus, FF factors cannot proxy
as DRF in the cross-section on CNAS stock market over the sample period.
168
General conclusions
The main objective of this dissertation is to explore the risk factors and factor models on
CNAS stock market based on the context of FF3F Model.
The main results of this dissertation are presented as follows:
- First of all, we re-examine the applicability of FF3F Model considering several

special features of Chinese stock market during July 2004 to May 2015. The
empirical results show that FF3F Model can explain a majority of time-series
variation of the CNAS stock returns during the research period, when using tradable
market value to weight the portfolios, total market capitalization to decide the size
breakpoint and B/P ratio instead of B/M ratio. The CSRs results are consistent with
most of the previous studies on Chinese stock market, market beta and SMB are
important determinants in explaining the cross-sectional variation in the average
stock returns over the sample period. There exists negative market premium and
positive size premium on CNAS stock market, however, we find no value premium
during the sample period. Those findings are robust with EIV adjustment and are
independent of research interval.
- We also investigate the applicability of the latest FF5F Model on CNAS stock
market during the period July 2010 to May 2015. To proceed with this examination,
we construct three sets of portfolios, six value-weighted Size-B/P portfolios, six
value-weighted Size-OP portfolios and six value-weighted Size-Inv portfolios. For
all the three sets of portfolios, the original three factors - market factor, size factor
and value factor – still have strong time-series explanatory power for the expected
excess returns in the presence of profitability and investment factors. There always
exists size effect in all three sets of portfolios and the excess returns are negatively
related to firm size; there exists value effect in SBP portfolios, profitability effect in
Size-OP portfolios and investment effect in Size-Inv portfolios. The explanatory
power of RMW factor only exists in the six Size-OP portfolios. CMA factor
explains the average return of portfolios that only in the extreme OP or Inv groups
(such as the weak OP group, robust OP group, the aggressive and conservative Inv
groups); while the significant loadings on CMA for the Size-B/P portfolios are
relatively dispersive.
- Comparing the performance of both FF3F Model and FF5F Model on CNAS stock
market, in the presence of profitability and investment factors, FF5F Model seems
169
General conclusions
not capture more variations of expected stock returns than the three-factor model
except the six value-weighted portfolios formed on size and operating profitability
(though the improvement is limited) on CNAS stock market during the research
period July 2010 to May 2015. However, the research period is relatively short in
this study, we suggest to apply the examination with longer time interval for the
FF5F Model on Chinese stock market in the future.
- We examine in chapter 2 whether FF factors SMB and HML proxy for the
innovations of selected state variables (aggregate dividend yield, one-month T-bill
rate, term spread and default spread) that describe future investment opportunities
on CNAS stock market during the period December 2006 to May 2015. Both time-
series and CSRs are performed on five comparative models. The empirical results
indicate that FF factors don’t lose their explanatory power with or without the
presence of the innovations of selected four state variables in both the time-series
and cross-sectional examinations on CNAS stock market over the research period.
Evidence from the CSRs also reveals that there are significant market risk premium
and size premium. We find the information contained in the innovation of aggregate
dividend yields (IDIV) seems totally captured by the combination of market beta
and size factor. FF factors might have played a limited role in capturing alternative
investment opportunities proxied by innovations of the selected four state variables.
- Since we find that FF factors cannot proxy for innovations of selected state
variables on CNAS stock market, for the sake of the meaning behind them, we
examine whether FF factors proxy for distress risk factor and compare whether
different methods of constructing factors result in the different outcomes. The
empirical results suggest that there is no significant evidence that FF factors are
proxying for distress risk on CNAS stock market. The presence of DRF has little
effect on the time-series explanatory power of FF three factors, instead, DRF can
explain partially the time-series excess average stock returns combining with FF
three factors. Comparing the TSR results by using two different methods as proxy
of distress risk, the distress risk factor constructed based on DLI seems performs
slightly better than that constructed based on O-score in capturing time-series
average returns. However, DRF is not an important determinant of cross-sectional
average returns, and FF factors cannot proxy as DRF in the cross-section on CNAS
stock market during our research period July 2005 to May 2015.
Moreover, our studies on the risk factors in China presented in this dissertation have also
provided new implications in practice on Chinese stock market:
170
General conclusions
First of all, considering several special features of Chinese stock market, our results
indicate that the market beta and size factor SMB are important determinants of cross-
sectional average stock returns. And the existence of size premium and lack of value
premium on CNAS stock market seem independent of the research period. Based on those
findings of FF3F Model on Chinese stock market, such as asset managers, they can build
portfolios that tilt towards the size factor SMB but not the value factor HML so that to gain
the size premium. Furthermore, asset managers or the individual investors are able to assess
the potential performance of a portfolio relative to the FF3F Model as a benchmark.
Then from the results of examination of FF5F Model on CNAS stock market, we conclude
that the profitability and investment factors have limited additional explanatory power, and
Fama-French Five-Factor Model does not have significant improvement in explaining
average excess stock returns comparing with the original three-factor model on CNAS
stock market, which is inconsistent with the findings on U.S. stock market. Similarly, if
investors want to invest in Chinese stock market, it’s better and easier to select the
portfolios that constructed based on FF3F instead of FF5F Model. However, if investors
invest on U.S. stock market, it is wiser to choose their portfolios that constructed based on
FF5F Model, since FF5F Model performs better than FF3F Model on U.S. stock market.
In addition, our attempts to find the economic explanation of FF factors on Chinese stock
market suggest that FF factors don’t lose their explanatory power in the presence of the
innovations of selected state variables in both time-series and CSRs on CNAS stock market
over the research period, and the presence of innovations of state variables do capture more
variation of average returns than original FF3F Model. The findings indicate that on
Chinese stock market, constructing portfolios tilt towards the innovations of the four
economic variables cannot gain extra risk premium in addition to the FF factors. FF3F
Model seems to be the best choice in practice so far on Chinese stock market.
Finally, our findings in the last chapter suggest that FF factors are not proxy of distress risk
on CNAS stock market, instead, the distress risk factor explains the time-series excess
average stock returns combining with FF three factors. The augmented four-factor model
explains the time-series variation of average excess stock returns slightly better than FF3F
Model on CNAS stock market. In this case, for example, if constructing portfolios tilt
towards our mimicking distress risk factor in addition to FF factors, one can expect more
average returns than constructing portfolios only based on the original FF factors. However,
the extra benefit comes from distress risk factor is limited. Moreover, for instance, our
findings are also can be implemented by companies to estimate the cost of equity; by
investors to evaluate the inherent value of equities and to make their decisions.
171
General conclusions
All in all, in order to keep up with the pace of China’s reforms and the globalization of its
economy, the Chinese stock market has experienced fast development and overall
institutional reforms. Considering the special features, the asset returns and its determinants
might be different between Chinese stock market and those of developed stock markets
such as U.S. and European markets. Our present research fails to find the economic
explanation for the success of FF factors on CNAS stock market, therefore, we propose to
consider the risk factors which feature the special characteristics of the Chinese stock
market or other economic variables that related to stock returns in further researches.
172
Bibliography
Abarbanell, J.S., and Bushee, B.J. (1998), Abnormal returns to a fundamental analysis
strategy, Accounting Review, 19–45.
Adams, G., McQueen, G.R., and Wood, R. (1999), The effects of inflation news on high
frequency stock returns.
Agarwal, V., and Taffler, R. (2008), Comparing the performance of market-based and
accounting-based bankruptcy prediction models, Journal of Banking & Finance, 32, 1541–
1551.
Aharoni, G., Grundy, B., and Zeng, Q. (2013), Stock returns and the Miller Modigliani
valuation formula: Revisiting the Fama French analysis, Journal of Financial Economics,
110, 347–357.
Ahlgren, N., Sjö, B., and Zhang, J. (2009), Panel cointegration of Chinese A and B shares,
Applied Financial Economics, 19, 1859–1871.
Alford, A.W., and Rakhlin, D.A. (2017), Implementing a Smart Beta Index: The
Implications of a Dual Performance Objective and Limited Liquidity, The Journal of
Portfolio Management, 43, 135–146.
Allen, L., and Saunders, A. (2002), Credit Risk Measurement. New Approaches to Value at
Risk and Other Paradigms, VJohn Wiley & Sons,.
Altman, E.I. (1968), Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy, The journal of finance, 23, 589–609.
Amanda, C., and Husodo, Z.A. (2014), Empirical Test of Fama French Three Factor Model
and Illiquidity Premium in Indonesia.
Amenc, N., Ducoulombier, F., Esakia, M., Goltz, F., and Sivasubramanian, S. (2017),
Accounting for Cross-Factor Interactions in Multifactor Portfolios without Sacrificing
Diversification and Risk Control, The Journal of Portfolio Management, 43, 99–114.
Anderson, C.W., and GARCIA-FEIJÓO, L. (2006), Empirical evidence on capital

investment, growth options, and security returns, The Journal of Finance, 61, 171–194.
Aretz, K., Bartram, S.M., and Pope, P.F. (2010), Macroeconomic risks and characteristic-
based factor models, Journal of Banking & Finance, 34, 1383–1399.
173
Bibliography
Aretz, K., Florackis, C., and Kostakis, A. (2014), Do Stock Returns Really Decrease with
Default Risk? New International Evidence, New International Evidence (September 8,
2014).
Asprem, M. (1989), Stock prices, asset portfolios and macroeconomic variables in ten
European countries, Journal of Banking & Finance, 13, 589–612.
Avramov, D., Chordia, T., Jostova, G., and Philipov, A. (2007), Momentum and credit
rating, The Journal of Finance, 62, 2503–2520.
Bahl, B. (2006), Testing the Fama and French Three-factor model and its variants for the
Indian stock returns.
Balakrishnan, A. (2014), Multifactor explanations of CAPM anomalies: An evidence for

Indian stock market, Asian Journal of Finance & Accounting, 6, 337–366.
Balakrishnan, A. (2016), Size, Value, and Momentum Effects in Stock Returns: Evidence
from India, Vision, 20, 1–8.
Banz, R.W. (1981), The relationship between return and market value of common stocks,
Journal of financial economics, 9, 3–18.
Bass, R., Gladstone, S., and Ang, A. (2017), Total Portfolio Factor, Not Just Asset,
Allocation, The Journal of Portfolio Management, 43, 38–53.
Basu, S. (1977), Investment performance of common stocks in relation to their price-

earnings ratios: A test of the efficient market hypothesis, The journal of Finance, 32, 663–
682.
Beaver, W.H. (1966), Financial ratios as predictors of failure, Journal of accounting

research, 71–111.
Beaver, W.H., McNichols, M.F., and Rhie, J.-W. (2005), Have financial statements become
less informative? Evidence from the ability of financial ratios to predict bankruptcy,
Review of Accounting Studies, 10, 93–122.
Beckers, S., Grinold, R., Rudd, A., and Stefek, D. (1992), The relative importance of
common factors across the European equity markets, Journal of Banking & Finance, 16,
75–95.
Begley, J., Ming, J., and Watts, S. (1996), Bankruptcy classification errors in the 1980s: An
empirical analysis of Altman’s and Ohlson’s models, Review of accounting Studies, 1,
267–284.
174
Bibliography
Bender, J., and Samanta, R. (2017), Quality Assurance: Demystifying the Quality Factor in
Equities and Bonds, The Journal of Portfolio Management, 43, 88–98.
Bhattacharjee, A., and Han, J. (2014), Financial distress of Chinese firms: Microeconomic,
macroeconomic and institutional influences, China Economic Review, 30, 244–262.
Black, F. (1972), Capital market equilibrium with restricted borrowing, The Journal of
Business, 45, 444–455.
Black, F., and Scholes, M. (1973), The pricing of options and corporate liabilities, The
journal of political economy, 637–654.
Boons, M. (2016), State variables, macroeconomic activity, and the cross section of
individual stocks, Journal of Financial Economics, 119, 489–511.
Bottazzi, L., and Corradi, V. (1991), Analysing the risk premium in the Italian stock market:
ARCH-M models versus non-parametric models, Applied Economics, 23, 535–541.
Breeden, D.T. (1979), An intertemporal asset pricing model with stochastic consumption
and investment opportunities, Journal of financial Economics, 7, 265–296.
Brennan, M.J., Wang, A.W., and Xia, Y. (2004), Estimation and test of a simple model of
intertemporal capital asset pricing, The Journal of Finance, 59, 1743–1776.
Brounen, D., de Jong, A., and Koedijk, K. (2004), Corporate finance in Europe: confronting
theory with practice, Financial Management, 33, 71–102.
Burmeister, E., Roll, R., and Ross, S.A. (1994), A practitioner’s guide to arbitrage pricing
theory.
Cakici, N., Chan, K., and Topyan, K. (2015), Cross-sectional stock return predictability in
China, The European Journal of Finance, 1–25.
Campbell, J.Y. (1987), Stock returns and the term structure, Journal of Financial
Economics, 18, 373–399.
Campbell, J.Y. (1991), A Variance Decomposition for Stock Returns, The Economic
Journal, 101, 157–179.
Campbell, J.Y. (1993), Intertemporal Asset Pricing without Consumption Data, The
American Economic Review, 487–512.
Campbell, J.Y. (1996), Understanding Risk and Return, Journal of Political Economy, 104,
298–345.
175
Bibliography
Campbell, J.Y., Hilscher, J., and Szilagyi, J. (2008), In search of distress risk, The Journal
of Finance, 63, 2899–2939.
Campbell, J.Y., Hilscher, J.D., and Szilagyi, J. (2011), Predicting financial distress and the
performance of distressed stocks.
Campbell, J.Y., and Shiller, R.J. (1988a), The dividend-price ratio and expectations of
future dividends and discount factors, Review of financial studies, 1, 195–228.
Campbell, J.Y., and Shiller, R.J. (1988b), Stock prices, earnings, and expected dividends,
The Journal of Finance, 43, 661–676.
Campbell, J.Y., and Vuolteenaho, T. (2004), Bad beta, good beta, The American Economic
Review, 94, 1249–1275.
Carhart, M.M. (1997), On Persistence in Mutual Fund Performance, The Journal of Finance,
52, 57–82.
Chan, A., and Chui, A.P. (1996), An Empirical Re-Examination of the Cross-Section of
Expected Returns: UK Evidence, Journal of Business Finance & Accounting, 23, 1435–
1452.
Chan, K., Chan, L.K., Jegadeesh, N., and Lakonishok, J. (2006), Earnings Quality and
Stock Returns, The Journal of Business, 79, 1041–1082.
Chan, K.C., Chen, N., and Hsieh, D.A. (1985), An exploratory investigation of the firm size
effect, Journal of Financial Economics, 14, 451–471.
Chan, K.C., and Chen, N.-F. (1991), Structural and return characteristics of small and large
firms, The Journal of Finance, 46, 1467–1484.
Charles, A., Darné, O., and Kim, J.H. (2016), International stock return predictability:
Evidence from new statistical tests, International Review of Financial Analysis.
Chava, S., and Jarrow, R.A. (2004), Bankruptcy prediction with industry effects, Review of
Finance, 8, 537–569.
Chava, S., and Purnanandam, A. (2010), Is Default Risk Negatively Related to Stock
Returns?, Review of Financial Studies, hhp107.
Chen, C., Hu, X., Shao, Y., and Wang, J. (2015), Fama-French in China: Size and Value
Factors in Chinese Stock Returns, University of Hong Kong Working Paper.
176
Bibliography
Chen, G.-M., Lee, B.-S., and Rui, O. (2001), Foreign ownership restrictions and market
segmentation in China’s stock markets, Journal of Financial Research, 24, 133–155.
Chen, J. (2002), Intertemporal CAPM and the cross-section of stock returns.
Chen, L., and Zhang, L. (2010), A better three-factor model that explains more anomalies,
Journal of Finance, 65, 563–595.
Chen, N.-F. (1991), Financial Investment Opportunities and the Macroeconomy, The
Chen, N.-F., Roll, R., and Ross, S.A. (1986), Economic forces and the stock market,
Journal of business, 383–403.
Chen, Xuanjuan, Kim, K.A., Yao, T., and Yu, T. (2010), On the predictability of Chinese
stock returns, Pacific-Basin Finance Journal, 18, 403–425.
Chen, Xiaohong, Wang, X., and Wu, D.D. (2010), Credit risk measurement and early
warning of SMEs: An empirical study of listed SMEs in China, Decision Support Systems,
49, 301–310.
Chen, Y., and Chu, G. (2014), Estimation of Default Risk Based on KMV Model—An
Empirical Study for Chinese Real Estate Companies, Journal of Financial Risk
Management, 2014.
Chen, Z. (2004), Cross-Sectional Variations and Three Factors Asset Pricing Model:
Empirical Evidences from China A Share Market [J], Chinese Journal of Management
Science, 6, 003.
Cheung, Y.-W., He, J., and Ng, L.K. (1997), What are the global sources of rational
variation in international equity returns?, Journal of International Money and Finance, 16,
821–836.
Cheung, Y.-W., and Ng, L.K. (1998), International evidence on the stock market and
aggregate economic activity, Journal of empirical finance, 5, 281–296.
Chiah, M., Chai, D., and Zhong, A. (2015), A better model? An empirical investigation of
the Fama-French five-factor model in Australia, in: 2015 Financial Markets & Corporate
Governance Conference.
Chui, A.C., and Wei, K.J. (1998), Book-to-market, firm size, and the turn-of-the-year effect:
Evidence from Pacific-Basin emerging markets, Pacific-Basin finance journal, 6, 275–293.
177
Bibliography
Claessens, S., Dasgupta, S., and Glen, J. (1995), Return behavior in emerging stock
markets, The World Bank Economic Review, 131–151.
Cochrane, J.H. (2005), Asset Pricing.
Cocoma, P., Czasonis, M., Kritzman, M., and Turkington, D. (2017), Facts about Factors,
The Journal of Portfolio Management, 43, 55–65.
Cohen, R.B., Gompers, P.A., and Vuolteenaho, T. (2002), Who underreacts to cash-flow
news? Evidence from trading between individuals and institutions, Journal of financial
Economics, 66, 409–462.
Connor, G. (1995), The Three Types of Factor Models: A Comparison of Their

Explanatory Power, Financial Analysts Journal, 51, 42–46.
Connor, G., and Sehgal, S. (2001), Tests of the Fama and French model in India.
Cooper, I., and Maio, P.F. (2016), Equity risk factors and the Intertemporal CAPM, Cross-
section of stock returns.
Cooper, M.J., Gulen, H., and Schill, M.J. (2008), Asset growth and the cross-section of
stock returns, The Journal of Finance, 63, 1609–1651.
Crosbie, P., and Bohn, J. (2003), Modeling default risk.
Daniel, K., and Titman, S. (1997), Evidence on the characteristics of cross sectional
variation in stock returns, The Journal of Finance, 52, 1–33.
Das, S. (2015), Empirical evidence of conditional asset pricing in the Indian stock market,
Economic Systems, 39, 225–239.
Dechow, P.M., Hutton, A.P., and Sloan, R.G. (2000), The relation between analysts’
forecasts of long-term earnings growth and stock price performance following equity
offerings, Contemporary Accounting Research, 17, 1–32.
Dichev, I.D. (1998), Is the risk of bankruptcy a systematic risk?, the Journal of Finance, 53,
1131–1147.
Dimson, E., Marsh, P., and Staunton, M. (2017), Factor-Based Investing: The Long-Term
Evidence, The Journal of Portfolio Management, 43, 15–37.
Dimson, E., and Mussavian, M. (1998), A brief history of market efficiency, European
financial management, 4, 91–103.
178
Bibliography
Docherty, P., Chan, H., and Easton, S. (2013), Can we treat empirical regularities as state
variables in the ICAPM? Evidence from Australia, Pacific-Basin Finance Journal, 22, 107–
124.
Dotsis, G. (2015), The Market Price of Risk of the Variance Term Structure, Journal of
Banking & Finance.
Drew, M. (2003), Beta, Firm Size, Book-to-Market Equity and Stock Returns, Journal of
the Asia Pacific Economy, 8, 354–379.
Drew, M.E., Naughton, T., and Veeraraghavan, M. (2003), Firm size, book-to-market
equity and security returns: Evidence from the Shanghai Stock Exchange, Australian
Journal of Management, 28, 119–139.
Drew, M.E., and Veeraraghavan, M. (2002), A closer look at the size and value premium in
emerging markets: Evidence from the Kuala Lumpur Stock Exchange, Asian Economic
Journal, 16, 337–351.
Duffie, D., and Singleton, K.J. (1999), Modeling term structures of defaultable bonds,
Review of Financial studies, 12, 687–720.
Elsas, R., El-Shaer, M., and Theissen, E. (2003), Beta and returns revisited: evidence from
the German stock market, Journal of International Financial Markets, Institutions and
Money, 13, 1–18.
Eun, C.S., and Huang, W. (2007), Asset pricing in China’s domestic stock markets: Is there
a logic?, Pacific-Basin Finance Journal, 15, 452–480.
Faff, R. (2001), An examination of the Fama and French three-factor model using
commercially available factors, Australian Journal of Management, 26, 1–17.
Fairfield, P.M., Whisenant, J.S., and Yohn, T.L. (2003), Accrued earnings and growth:
Implications for future profitability and market mispricing, The accounting review, 78,
353–371.
Fama, E.F. (1981), Stock returns, real activity, inflation, and money, The American
Economic Review, 71, 545–565.
Fama, E.F. (1990), Stock returns, expected returns, and real activity, The Journal of
Finance, 45, 1089–1108.
Fama, E.F. (1991), Efficient capital markets: II, The journal of finance, 46, 1575–1617.
179
Bibliography
Fama, E.F., and French, K.R. (1988), Dividend yields and expected stock returns, Journal
of Financial Economics, 22, 3–25.
Fama, E.F., and French, K.R. (1989), Business conditions and expected returns on stocks
and bonds, Journal of financial economics, 25, 23–49.
Fama, E.F., and French, K.R. (1992), The cross-section of expected stock returns, the
Fama, E.F., and French, K.R. (1993), Common risk factors in the returns on stocks and
bonds, Journal of financial economics, 33, 3–56.
Fama, E.F., and French, K.R. (1995), Size and book-to-market factors in earnings and
returns, The Journal of Finance, 50, 131–155.
Fama, E.F., and French, K.R. (1996), Multifactor explanations of asset pricing anomalies,
The journal of finance, 51, 55–84.
Fama, E.F., and French, K.R. (1998), Value versus growth: The international evidence, The
journal of finance, 53, 1975–1999.
Fama, E.F., and French, K.R. (2006), Profitability, investment and average returns, Journal
of Financial Economics, 82, 491–518.
Fama, E.F., and French, K.R. (2008), Dissecting anomalies, The Journal of Finance, 63,
1653–1678.
Fama, E.F., and French, K.R. (2015a), A five-factor asset pricing model, Journal of
Financial Economics, 116, 1–22.
Fama, E.F., and French, K.R. (2015b), International tests of a five-factor asset pricing
model, Fama-Miller Working Paper.
Fama, E.F., and French, K.R. (2017), International tests of a five-factor asset pricing model,
Journal of Financial Economics, 123, 441–463.
Fama, E.F., and MacBeth, J.D. (1973), Risk, return, and equilibrium: Empirical tests, The
journal of political economy, 607–636.
Fama, E.F., and Schwert, G.W. (1977), Asset returns and inflation, Journal of financial
economics, 5, 115–146.
180
Bibliography
Ferdian, I.R., Omar, M.A., and Dewi, M.K. (2011), Firm Size, Book to Market Equity, and
Security Returns: Evidence from the Indonesian Shariah Stocks, Journal of Islamic
Economics, Banking and Finance, 7.
Fernald, J., and Rogers, J.H. (2002), Puzzles in the Chinese stock market, Review of
Economics and Statistics, 84, 416–432.
Ferson, W.E., and Harvey, C.R. (1991), The variation of economic risk premiums, Journal
of Political Economy, 99, 385–415.
Ferson, W.E., and Harvey, C.R. (1993), The risk and predictability of international equity
returns, Review of financial Studies, 6, 527–566.
FitzPatrick, P.J. (1932), A comparison of the ratios of successful industrial enterprises with
those of failed companies.
Frankel, R., and Lee, C.M. (1998), Accounting valuation, market expectation, and cross-
sectional stock returns, Journal of Accounting and economics, 25, 283–319.
Friewald, N., Wagner, C., and Zechner, J. (2014), The Cross-Section of Credit Risk Premia
and Equity Returns, The Journal of Finance, 69, 2419–2469.
Fung, H.-G., Lee, W., and Leung, W.K. (2000), SEGMENTATION OF THE A-AND B-
SHARE CHINESE EQUITY MARKETS, Journal of Financial Research, 23, 179–195.
Gan, C., Hu, B., Liu, Y., and Li, Z. (2013), An empirical cross-section analysis of stock
returns on the Chinese A-share stock market.
Garlappi, L., Shu, T., and Yan, H. (2008), Default risk, shareholder advantage, and stock
returns, Review of Financial Studies, 21, 2743–2778.
Garlappi, L., and Yan, H. (2011), Financial distress and the cross-section of equity returns,
Gaunt, C. (2004), Size and book to market effects and the Fama French three factor asset
pricing model: evidence from the Australian stockmarket, Accounting & Finance, 44, 27–
44.
Gaur, J., Dash, M., and others (2015), Macroeconomic Factors and Performance of Indian
Stock Market, Journal of Applied Management and Investments, 4, 11–15.
Geng, R., Bose, I., and Chen, X. (2015), Prediction of financial distress: An empirical study
of listed Chinese companies using data mining, European Journal of Operational Research,
241, 236–247.
181
Bibliography
George, T.J., and Hwang, C.-Y. (2010), A resolution of the distress risk and leverage
puzzles in the cross section of stock returns, Journal of Financial Economics, 96, 56–79.
Geske, R., and Roll, R. (1983), The fiscal and monetary linkage between stock returns and
inflation, The Journal of Finance, 38, 1–33.
Gharghori, P., Chan, H., and Faff, R. (2006), Investigating the performance of alternative
default-risk models: Option-based versus accounting-based approaches, Australian Journal
of Management, 31, 207–234.
Gharghori, P., Chan, H., and Faff, R. (2007), Are the Fama-French factors proxying default
risk?, Australian Journal of Management, 32, 223–249.
Gharghori, P., Chan, H., and Faff, R. (2009), Default risk and equity returns: Australian
evidence, Pacific-Basin Finance Journal, 17, 580–593.
Gibbons, M.R., Ross, S.A., and Shanken, J. (1989), A test of the efficiency of a given
portfolio, Econometrica: Journal of the Econometric Society, 1121–1152.
Graham, J.R., and Harvey, C.R. (2001), The theory and practice of corporate finance:
Evidence from the field, Journal of financial economics, 60, 187–243.
Grice, J.S., and Dugan, M.T. (2001), The limitations of bankruptcy prediction models:
Some cautions for the researcher, Review of Quantitative Finance and Accounting, 17,
151–166.
Griffin, J.M., and Lemmon, M.L. (2002), Book–to–market equity, distress risk, and stock
returns, The Journal of Finance, 57, 2317–2336.
Hahn, J., and Lee, H. (2001), An Empirical Investigation of Risk and Return under Capital
Market Imperfections, Available at SSRN 283368.
Hahn, J., and Lee, H. (2006), Yield spreads as alternative risk factors for size and book-to-
market, Journal of Financial and Quantitative Analysis, 41, 245–269.
Hanhardt, A., and Ansotegui Olcoz, C. (2008), Do the Fama and French Factors Proxy for
State Variables that Predict Macroeconomic Growth in the Eurozone?
Hansen, L.P., Heaton, J., and Li, N. (2005), Consumption strikes back?: Measuring long-
run risk.
Harvey, C.R. (1989), Time-varying conditional covariances in tests of asset pricing models,
182
Bibliography
Haugen, R.A., and Baker, N.L. (1996a), Commonality in the determinants of expected
stock returns, Journal of Financial Economics, 41, 401–439.
Haugen, R.A., and Baker, N.L. (1996b), Commonality in the determinants of expected
stock returns, Journal of Financial Economics, 41, 401–439.
Hillegeist, S.A., Keating, E.K., Cram, D.P., and Lundstedt, K.G. (2004), Assessing the
probability of bankruptcy, Review of accounting studies, 9, 5–34.
Ho, R.Y., Strange, R., and Piesse, J. (2006), On the conditional pricing effects of beta, size,
and book-to-market equity in the Hong Kong market, Journal of International Financial
Markets, Institutions and Money, 16, 199–214.
Homsud, N., Wasunsakul, J., Phuangnark, S., and Joongpong, J. (2009), A study of Fama
and French three factors model and capital asset pricing model in the Stock exchange of
Thailand, International Research Journal of Finance and Economics, 25, 31–40.
Hosseini, S.M., Ahmad, Z., and Lai, Y.W. (2011), The role of macroeconomic variables on
stock market index in China and India, International Journal of Economics and Finance, 3,
233.
Hou, K., Xue, C., and Zhang, L. (2014), Digesting anomalies: An investment approach,
Review of Financial Studies.
Hu, J. (1999), The evolution of China stock market and its institutions, Economic Science,
Beijing.
Huang, Y., Yang, J., and Zhang, Y. (2013), Value premium in the Chinese stock market:
free lunch or paid lunch?, Applied Financial Economics, 23, 315–324.
Hung, C.-H.D., Chen, Q., and Fang, V. (2015), Non-Tradable Share Reform, Liquidity, and
Stock Returns in China, International Review of Finance, 15, 27–54.
Hung, D.C.-H., Shackleton, M., and Xu, X. (2004), CAPM, Higher Co-moment and Factor
Models of UK Stock Returns, Journal of Business Finance & Accounting, 31, 87–112.
In, F., and Kim, S. (2007), A note on the relationship between Fama–French risk factors
and innovations of ICAPM state variables, Finance Research Letters, 4, 165–171.
Isakov, D. (1999), Is beta still alive? Conclusive evidence from the Swiss stock market, The
European Journal of Finance, 5, 202–212.
Jagannathan, R., and Wang, Z. (1996), The conditional CAPM and the cross-section of
expected returns, The Journal of finance, 51, 3–53.
183
Bibliography
Jarrow, R.A., and Turnbull, S.M. (1995), Pricing derivatives on financial securities subject
to credit risk, The journal of finance, 50, 53–85.
Jensen, M.C., Black, F., and Scholes, M.S. (1972), The capital asset pricing model: Some
empirical tests.
Kang, J., Liu, M.-H., and Ni, S.X. (2002), Contrarian and momentum strategies in the
China stock market: 1993–2000, Pacific-Basin Finance Journal, 10, 243–265.
Keim, D.B., and Stambaugh, R.F. (1986), Predicting returns in the stock and bond markets,
Journal of financial Economics, 17, 357–390.
Khan, M. (2008), Are accruals mispriced? Evidence from tests of an Intertemporal Capital
Asset Pricing Model, Journal of Accounting and Economics, 45, 55–77.
Kim, J.H., Kim, W.C., and Fabozzi, F.J. (2017), Robust Factor-Based Investing, The
Journal of Portfolio Management, 43, 157–164.
Kleinert, M.K. (2014), Comparison of bankruptcy prediction models of Altman (1969),

Ohlson (1980) and Zmijewski (1984) on German and Belgian listed companies between
2008-2013.
Kothari, S.P., and Shanken, J. (1997), Book-to-market, dividend yield, and expected market
returns: A time-series analysis, Journal of Financial Economics, 44, 169–203.
Kothari, S.P., Shanken, J., and Sloan, R.G. (1995a), Another look at the cross-section of
expected stock returns, The Journal of Finance, 50, 185–224.
Kothari, S.P., Shanken, J., and Sloan, R.G. (1995b), Another look at the cross-section of
expected stock returns, The Journal of Finance, 50, 185–224.
Lakonishok, J., Shleifer, A., and Vishny, R.W. (1994a), Contrarian investment,
extrapolation, and risk, The journal of finance, 49, 1541–1578.
Lakonishok, J., Shleifer, A., and Vishny, R.W. (1994b), Contrarian investment,
extrapolation, and risk, The journal of finance, 49, 1541–1578.
Lam, K.S. (2002), The relationship between size, book-to-market equity ratio, earnings–
price ratio, and return for the Hong Kong stock market, Global Finance Journal, 13, 163–
179.
Lau, S.T., Lee, C.T., and McInish, T.H. (2002), Stock returns and beta, firms size, E/P,
CF/P, book-to-market, and sales growth: evidence from Singapore and Malaysia, Journal of
multinational financial management, 12, 207–222.
184
Bibliography
Law, D., and Roache, S.K. (2015), Assessing Default Risks for Chinese Firms: A Lost
Cause?, SSRN Scholarly Paper No. ID 2633926.
Lev, B., and Zarowin, P. (1999), The Boundaries of Financial Reporting and How to
Extend Them (Digest Summary), Journal of Accounting research, 37, 353–385.
Lewellen, J., Nagel, S., and Shanken, J. (2010), A skeptical appraisal of asset pricing tests,
Journal of Financial economics, 96, 175–194.
Li, Y. (1997), Intertemporal Asset Pricing Without Consumption Data: Empirical Tests,
Journal of Financial Research, 20, 53–69.
Liang, P. (2013), Decoupled or Not? What Drives Chinese Stock Markets: Domestic or
Global Factors?, Journal of Accounting and Finance, 13, 40–54.
Liang, P., and Willett, T.D. (2015), Chinese stocks during 2000–2013: Bubbles and busts
or fundamentals?, The Chinese Economy, 48, 199–214.
Liang, X. (2012), An empirical estimation of the default risk of Chinese listed company
based on the Merton-KMV Model.
Liao, L., and Shen, H.B. (2008), Fama-French Three Factors Model and the Effect of the
Split-share Structure Reform, The Journal of Quantitative and Technical Economics, 9,
117–125.
Liew, J., and Vassalou, M. (2000), Can book-to-market, size and momentum be risk factors
that predict economic growth?, Journal of Financial Economics, 57, 221–245.
Lilti, J.-J., and Montagner, H.R.-L. (1998), Beta, size and returns: a study on the French
Stock Exchange, Applied Financial Economics, 8, 13–20.
Lin, H., and Chen, X. p (2008), Is default risk a systematic risk of Chinese stock markets?,
in: 2008 International Conference on Management Science and Engineering 15th Annual
Conference Proceedings, 1169–1174.
Lintner, J. (1965), The valuation of risk assets and the selection of risky investments in
stock portfolios and capital budgets, The review of economics and statistics, 13–37.
Lioui, A., and Poncet, P. (2011), Misunderstanding risk and return?, Finance, 32, 91–136.
Litterman, R.B., and Scheinkman, J. (1991), Common factors affecting bond returns, The
Journal of Fixed Income, 1, 54–61.
185
Bibliography
Litzenberger, R.H., and Ramaswamy, K. (1979), The effect of personal taxes and dividends
on capital asset prices: Theory and empirical evidence, Journal of financial economics, 7,
163–195.
Liu, D. (2017), Pure Quintile Portfolios, The Journal of Portfolio Management, 43, 115–
129.
Liu, G.-M., and Yang, C. (2010), Application of Fama-French multi-factor model in

China’s bond market during recent financial crisis, Journal of Zhejiang University (Science
Edition), 37, 396–400.
Liu, Z.J., and Wang, Y.S. (2016), Corporate Failure Prediction Models for Advanced
Research in China: Identifying the Optimal Cut Off Point, Asian Economic and Financial
Review, 6, 54.
Lu, C., Li, X. w, and Pan, H. b (2006), Empirical Research on Credit Condition of Chinese
Listed Corporation with KMV Default Model, in: 2006 International Conference on
Machine Learning and Cybernetics, 2387–2390.
Lyandres, E., Sun, L., and Zhang, L. (2008), The new issues puzzle: Testing the
investment-based explanation, Review of Financial Studies, 21, 2825–2855.
MacKinlay, A.C. (1995a), Multifactor models do not explain deviations from the CAPM,
MacKinlay, A.C. (1995b), Multifactor models do not explain deviations from the CAPM,
Mao, X., Chen, M., and Yang (2008), Long-run Return Performance Following Listed
Rights Issue: Based on the Improved Three-factor Model, Journal of Financial Research, 5,
013.
Markowitz, H. (1952), Portfolio selection, The journal of finance, 7, 77–91.
Markowitz, H. (1959), Portfolio Selection, Efficent Diversification of Investments, J. Wiley.
Maroney, N., and Protopapadakis, A. (2002), The book-to-market and size effects in a
general asset pricing model: Evidence from seven national markets, European Finance
Review, 6, 189–221.
Martinsa, C.C., and Eid Jr, W. (2015), Pricing Assets with Fama and French 5–Factor
Model: a Brazilian market novelty, in: XV Encontro Brasileiro de Finanças.
McFadden, D., and others (1973), Conditional logit analysis of qualitative choice behavior.
186
Bibliography
McKean, H.P. (1969), Stochastic integrals, American Mathematical Soc.
Mensah, Y.M. (1984), An examination of the stationarity of multivariate bankruptcy

prediction models: A methodological study, Journal of Accounting Research, 380–395.
Merton, R.C. (1973a), An intertemporal capital asset pricing model, Econometrica: Journal
of the Econometric Society, 867–887.
Merton, R.C. (1973b), Theory of rational option pricing, The Bell Journal of economics
and management science, 141–183.
Merton, R.C. (1974), On the pricing of corporate debt: The risk structure of interest rates,
The Journal of finance, 29, 449–470.
Morelli, D. (2007), Beta, size, book-to-market equity and returns: A study based on UK
data, Journal of Multinational Financial Management, 17, 257–272.
Morelli, D. (2012), Security returns, beta, size, and book-to-market equity: evidence from
the Shanghai A-share market, Review of Quantitative Finance and Accounting, 38, 47–60.
Mu, T. (2016), A Study on Dynamic Relationship between Macroeconomic Variables and

Stock Markets in the United States, Germany, and Hong Kong, Browser Download This
Paper.
Mullins, M., and Wadhwani, S.B. (1989), The effect of the stock market on investment: A
comparative study, European Economic Review, 33, 939–956.
Ni, J., Kwak, W., Cheng, X., and Gong, G. (2014), The determinants of bankruptcy for
Chinese firms, Review of Pacific Basin Financial Markets and Policies, 17, 1450012.
Novy-Marx, R. (2013), The other side of value: The gross profitability premium, Journal of
Financial Economics, 108, 1–28.
Ohlson, J.A. (1980), Financial ratios and the probabilistic prediction of bankruptcy, Journal
of accounting research, 109–131.
Outecheva, N. (2007), Corporate financial distress: An empirical analysis of distress risk,

University of St. Gallen.
Parker, J.A., and Julliard, C. (2005), Consumption risk and the cross section of expected
returns, Journal of Political Economy, 113, 185–222.
Petkova, R. (2006), Do the Fama–French factors proxy for innovations in predictive

variables?, The Journal of Finance, 61, 581–612.
187
Bibliography
Petkova, R., and Zhang, L. (2005), Is value riskier than growth?, Journal of Financial
Economics, 78, 187–202.
Phong, N.A., and Hoang, T.V. (2012), Applying Fama and French Three Factors Model
and Capital Asset Pricing Model in the Stock Exchange of Vietnam, International Research
Journal of Finance and Economics.
Podkaminer, E. (2017), Smart Beta is the Gateway Drug to Risk Factor Investing, The
Journal of Portfolio Management, 43, 130–134.
Pongsatat, S., Ramage, J., and Lawrence, H. (2004), Bankruptcy prediction for large and
small firms in Asia: a comparison of Ohlson and Altman, Journal of Accounting and
Corporate Governance, 1, 1–13.
Pramod Kumar, N., and Puja, P. (2012), The impact of macroeconomic fundamentals on
stock prices revisited: An evidence from Indian data.
Ranjan Dash, S., and Mahakud, J. (2013), Conditional multifactor asset pricing model and
market anomalies, Journal of Indian Business Research, 5, 271–294.
Reisz, A.S., and Perlich, C. (2007), A market-based framework for bankruptcy prediction,
Journal of Financial Stability, 3, 85–131.
Richardson, S.A., and Sloan, R.G. (2003), External financing and future stock returns,
Rodney L. White Center for Financial Research Working Paper.
Richardson, S.A., Sloan, R.G., Soliman, M.T., and Tuna, I. (2005), Accrual reliability,
earnings persistence and stock prices, Journal of accounting and economics, 39, 437–485.
Richardson, S.A., Sloan, R.G., Soliman, M.T., and Tuna, I. (2006), The implications of
accounting distortions and growth for accruals and profitability, The Accounting Review,
81, 713–743.
Roll, R. (1977), A critique of the asset pricing theory’s tests Part I: On past and potential
testability of the theory, Journal of financial economics, 4, 129–176.
Roll, R., and Ross, S.A. (1980), An empirical investigation of the arbitrage pricing theory,
Ross, S.A. (1976), The arbitrage theory of capital asset pricing, Journal of economic theory,
13, 341–360.
Ross, S.A., Westerfield, R.W., and Jaffe, J.F. (1998), Corporate Finance, Homewood, IL,
McGraw-Hill/Irwin. Previous editions, 1993, 1999.
188
Bibliography
Rozeff, M.S. (1984), Dividend yields are equity risk premiums, The Journal of Portfolio
Management, 11, 68–75.
Rutledge, R.W., Zhang, Z., and Karim, K. (2008), Is there a size effect in the pricing of
stocks in the chinese stock markets?: The case of bull versus bear markets, Asia-Pacific
Financial Markets, 15, 117–133.
Schwert, G.W. (1990), Stock returns and real activity: A century of evidence, The Journal
of Finance, 45, 1237–1257.
Sehgal, S., and Balakrishnan, A. (2013), Robustness of fama-french three factor model:
further evidence for indian stock market, Vision, 17, 119–127.
Sehgal, S., Subramaniam, S., and Deisting, F. (2014), Tests of equity market anomalies for
select emerging markets.
Shanken, J. (1992), On the estimation of beta-pricing models, Review of Financial studies,

5, 1–33.
Shanken, J., and Weinstein, M.I. (2006), Economic forces and the stock market revisited,
Journal of Empirical Finance, 13, 129–144.
Sharpe, W.F. (1964), Capital asset prices: A theory of market equilibrium under conditions
of risk, The journal of finance, 19, 425–442.
Shi, Q., Li, B., and Chung, R. (2015), Augmenting the Intertemporal CAPM with Inflation:
Some Further Evidence in Alternative Models.
Shum, W.C., and Tang, G.Y. (2005), Common risk factors in returns in Asian emerging
stock markets, International Business Review, 14, 695–717.
Shumway, T. (2001), Forecasting bankruptcy more accurately: A simple hazard model*,

The Journal of Business, 74, 101–124.
Simpson, M.W., and Ramchander, S. (2008), An inquiry into the economic fundamentals of
the Fama and French equity factors, Journal of Empirical Finance, 15, 801–815.
Siti Noorahayusolah, K. (2011), The Impact of Macroeconomic Forces on Stock Returns in

Malaysia, Universiti Utara Malaysia.
Sloan, R. (1996), Do stock prices fully reflect information in accruals and cash flows about
future earnings?(Digest summary), Accounting review, 71, 289–315.
189
Bibliography
Stock, J.H., and Watson, M.W. (1989), New indexes of coincident and leading economic
indicators, NBER macroeconomics annual, 4, 351–394.
Sudiyatno, B., and Irsad, M. (2013), STUDY OF THREE FACTORS MODEL FAMA
AND FRENCH IN INDONESIA STOCK EXCHANGE (Study on the Stock LQ 45),
Sustainable Competitive Advantage (SCA), 2.
Sun, Q., and Tong, W.H. (2000), The effect of market segmentation on stock prices: The
China syndrome, Journal of Banking & Finance, 24, 1875–1902.
Taneja, Y.P. (2010), Revisiting fama french three-factor model in indian stock market,
Vision: The Journal of Business Perspective, 14, 267–274.
Titman, S., Wei, K.J., and Xie, F. (2004), Capital investments and stock returns, Journal of
financial and Quantitative Analysis, 39, 677–700.
Tripathi, V., and Seth, R. (2014), Stock market performance and macroeconomic factors:
The study of Indian equity market, Global Business Review, 15, 291–316.
Vasicek, O.A. (1984), Credit valuation, KMV Corporation, March.
Vassalou, M. (2003), News related to future GDP growth as a risk factor in equity returns,
Journal of financial economics, 68, 47–73.
Vassalou, M., and Xing, Y. (2004), Default Risk in Equity Returns, The Journal of Finance,
59, 831–868.
Wang, C. (2004), Relative strength strategies in China’s stock market: 1994–2000, Pacific-
Basin Finance Journal, 12, 159–177.
Wang, F., and Xu, Y. (2004), What Determines Chinese Stock Returns?, Financial Analysts
Journal, 60, 65–77.
Wang, Y., and Campbell, M. (2010a), Financial ratios and the prediction of bankruptcy: the
Ohlson model applied to Chinese publicly traded companies, the Journal of organizational
leadership & business, 1–15.
Wang, Y., and Campbell, M. (2010b), Business failure prediction for publicly listed
companies in China, Journal of Business and Management, 16, 75.
Wang, Y., and Di Iorio, A. (2007), The cross section of expected stock returns in the
Chinese A-share market, Global Finance Journal, 17, 335–349.
190
Bibliography
Wong, K.A., Tan, R.S.K., and Liu, W. (2006), The cross-section of stock returns on the
Shanghai Stock Exchange, Review of Quantitative Finance and Accounting, 26, 23–39.
Wu, H. (2011), The value and size effect—are there firm-specific-risks in China’s domestic
stock markets?, International Journal of Economics and Finance, 3, 26.
Wu, Y., Gaunt, C., and Gray, S. (2010), A comparison of alternative bankruptcy prediction
models, Journal of Contemporary Accounting & Economics, 6, 34–45.
Xie, H. (2001), The mispricing of abnormal accruals, The accounting review, 76, 357–373.
Xing, Y. (2008a), Interpreting the Value Effect Through the Q-Theory: An Empirical
Investigation, The Review of Financial Studies, 21, 1767–1795.
Xing, Y. (2008b), Interpreting the value effect through the Q-theory: An empirical
investigation, Review of Financial Studies, 21, 1767–1795.
Zhang, S., and Xu, J. (2013), The fama-french three factors in the Chinese stock market.
Zhou, Z., and YANG, Y. (2007), Using the Distance to Default to Identify Credit Risks: An
Empirical Study [J], Audit & Economy Research, 2, 016.
Zivot, E., and Wang, J. (2007), Modeling financial time series with S-Plus®, Springer
Science & Business Media.
191
Appendix A
Errors-in-Variables (EIV) problem and Shanken

correction procedure
In applying standard OLS method to a CSR, we assume that the right-hand variables β are
fixed. However, when implementing Fama-MacBeth two-stage approach, the β in the CSR
are not fixed, of course, but are estimated in the TSR. Therefore, the explanatory variable in
the CSR is measured with error. Hence, it is very important to determine whether the
relation between average stock returns and the risk factors is the result of the
misspecification of the asset pricing model or is simply a consequence of the EIV problem.
Shanken (1992) modifies the traditional two-pass procedure and derives an asymptotic
distribution of the CSR estimator within a multifactor framework in which asset returns are
generated by portfolio returns and prespecified factors. Shanken also provides an
adjustment for the standard errors of the CSR estimators with a multifactor interpretation
Cochrane (2005) derives the correct asymptotic standard errors due to Shanken (1992),.
With the simplifying assumption that the errors ε are i.i.d. over time and independent of the
factors, the result is
  1
    
1 1
 2 ˆ     
T
      1       
1
f f 

(A.1)
where λ is the factor risk premia, β is the TSR coefficients matrix,  is the variance-
covariance matrix of residuals,  f is the variance-covariance matrix of the factors.
193
Appendix B
Performance of FF3F Model and FF5F Model on U.S.

stock market
Table B.1 Time-series regression of six value-weighted Size-B/M portfolios on U.S. stock
market (period: July 2004- May 2015, 131 months)
The time-series regression results of six value-weighted Size-B/M portfolios on FF3F Model are
displayed in this table. Across the columns are the two size groups (Small and Big) and across the
rows are the three B/M ratio groups (Low, Medium and High). The left part of the table reports the
coefficients obtained from the time-series regressions and adjusted R-square. Correspondingly, the
right part of the table is t-statistics corrected for heteroscedasticity and autocorrelation using the
Newey-West estimator, and the standard error of the estimation. Numbers in bold are the t-statistics
which are significant at 5% confidence level.
L M H L M H
a t (a )
S -0.1217 0.1085 0.0093 -1.7229 2.4742 0.2413
B 0.0851 -0.0443 -0.0444 2.1144 -0.6264 -0.5223
b t (b)
S 1.0754 0.9785 0.9863 41.1094 69.4224 84.3440
B 0.9633 1.0243 1.0523 58.2424 40.0675 38.0928
s t (s)
S 1.0115 0.8834 0.8824 24.8082 38.7934 54.5879
B -0.1132 -0.1264 0.0162 -4.5531 -2.8759 0.3696
h t (h)
S -0.2140 0.2183 0.6748 -4.5584 8.0641 30.4628
B -0.2679 0.1040 0.8434 -11.5717 2.2594 12.6276
S 0.9781 0.9894 0.9945 0.0086 0.0057 0.0045
B 0.9852 0.9691 0.9683 0.0047 0.0078 0.0100
195
196
Table B.2 Time-series regressions of 25 value-weighted Size-B/M portfolios on FF3F Model, U.S. stock market (July 2004- May
2015; 131 months)
Appendix B
This table presents the time-series regressions results of 25 value-weighted Size-B/M portfolios on FF3F Model, Across the columns are five
size groups and across the rows are five B/M groups. The left part of the table is the coefficients of the regressions and adjusted R-square.
Correspondingly, the right part of the table is t-statistics corrected for heteroscedasticity and autocorrelation using the Newey-West estimator,
and the standard error of the residuals. Numbers in bold are the t-statistics which are significant at 5% confidence level.
Size B/P ratio
L 2 3 4 H L 2 3 4 H
a t (a )
S -0.6227 -0.0461 -0.2137 -0.0201 0.0095 -3.8594 -0.3958 -2.2758 -0.1898 0.0700
2 -0.0157 0.1969 0.2419 0.0694 -0.1141 -0.1646 2.0533 2.4736 0.5674 -1.1176
3 -0.0001 0.2748 0.2386 0.2086 0.1795 -0.0012 2.3832 2.0916 1.7311 1.1510
4 0.2159 0.1115 -0.2138 0.1989 -0.0792 1.9938 1.0221 -1.1539 1.7058 -0.4470
B 0.0118 0.1418 0.1174 -0.4563 0.0876 0.1585 1.8226 1.2430 -2.1494 0.3999
b t (b)
S 1.1111 0.9998 0.9807 0.8738 0.9491 21.3603 26.4149 40.2759 18.7566 23.2255
2 1.0783 0.9970 0.9603 0.9170 1.0880 30.6065 34.4155 35.9649 38.7462 32.5759
3 1.1181 1.0611 1.0178 1.0175 0.9960 24.4732 38.8964 28.8771 31.6152 25.8979
4 1.0636 1.0948 1.1823 1.0606 1.0737 23.9754 25.6645 15.7197 18.5949 22.3479
B 0.9766 0.9204 0.9536 1.1195 1.0956 32.9666 51.0610 26.0271 12.5321 13.9923
s t (s)
S 1.2024 1.1139 0.9662 0.9804 0.9628 15.0696 19.5260 18.0505 24.1812 15.0989
2 1.0246 1.0148 0.9695 0.8377 0.8737 14.0170 30.4977 20.2644 24.1121 13.6579
3 0.7051 0.6803 0.6301 0.5290 0.6720 12.6354 13.1719 13.3856 8.0100 10.1764
4 0.4484 0.3314 0.3331 0.2586 0.2551 10.0698 5.9816 4.1488 3.6079 3.6529
B -0.1985 -0.1845 -0.1960 -0.3180 0.0284 -4.6713 -5.0276 -3.5593 -3.8461 0.2377
Appendix B
Table B.2 Continued
Size B/P ratio

L 2 3 4 H L 2 3 4 H
h t (h)
S -0.1231 0.0783 0.2095 0.4896 0.7294 -1.9032 1.5451 4.3873 6.2762 15.9565
2 -0.2822 -0.0499 0.2415 0.4500 0.8687 -3.5719 -0.9898 7.2889 11.6221 14.8860
3 -0.3366 -0.0488 0.0939 0.2361 0.6287 -8.7449 -0.8861 2.3422 3.7369 7.6054
4 -0.3117 -0.1128 -0.0435 0.2368 0.6970 -7.4663 -1.7359 -0.3747 2.4142 6.8010
B -0.3729 0.0392 0.1710 0.5757 0.8925 -9.0766 1.0111 3.6067 7.2239 7.8523
S 0.9122 0.9398 0.9606 0.955 0.9655 0.0194 0.0147 0.0113 0.0119 0.0115
2 0.9454 0.9602 0.9655 0.9589 0.9653 0.0138 0.0113 0.0105 0.0112 0.0127
3 0.9452 0.9500 0.9457 0.9283 0.9114 0.0130 0.0121 0.0124 0.0143 0.0178
4 0.9460 0.9397 0.9196 0.9102 0.9231 0.0116 0.0125 0.0158 0.0157 0.0163
B 0.9595 0.9429 0.9248 0.9155 0.8728 0.0078 0.0094 0.0116 0.0157 0.0222
197
Appendix B
Table B.3 Cross-sectional regressions on FF3F Model of six Size-B/M portfolios and 25 Size-
B/M portfolios, U.S. stock market (July 2004- May 2015)
This table presents the results of cross-section regressions on FF3F Model of FF six value-weighted
Size-B/M portfolios (Panel A) and 25 value-weighted Size-B/M portfolios (Panel B). In each panel,
the first row is the cross-sectional regressions’ coefficients (coef.); the second row is the
corresponding Fama-MacBeth t-statistics (FM t-stats) at 5% confidence level, and the third row is
the Shanken corrected t-statistics (SH t-stats). The numbers in bold are the t-stats which are
significant at 5% level. The adjusted R-squares are percentage values.
Regression: Ri ,t  R f  i   M bî   SMB ˆsi   HML hî  i ,t
 M  SMB  HML Adj. R 2 (%)
Panel A: Cross-sectional regression of six Size-B/M portfolios
gamma (coef.) 0.0185 -0.0117 0.0012 0.0002 58.14
FM t-stats (2.0663) (-1.2023) (0.6239) (0.1026)
SH t-stats (-3.1339) (0.6235) (0.1025)
Panel B: Cross-sectional regression of 25 Size-B/M portfolios
gamma (coef.) 0.0132 -0.5547 0.0561 -0.0730 40.76
FM t-stats (3.0609) (-0.9691) (0.2817) (-0.3246)
SH t-stats (-1.4809) (0.2813) (-0.3232)
Appendix B
Table B.4 Time-series regressions of six value-weighted Size-B/M portfolios, Size-OP

portfolios and Size-Inv portfolios on FF5F Model on U.S. stock market (July 2010 to May
2015, 59 months)
This table presents the time-series regressions results of FF5F model. In each panel, the regression
intercept a , the regression coefficients b , s , h , r and c of market factor, size factor, value factor,
profitability factor and investment factor, adjusted R square are respectively presented in the left
part of the table, the corresponding t-statistics corrected for heteroscedasticity and autocorrelation
using the Newey-West estimator and residual standard error are presented in the right part. Panel A
is the regressions on six Size-B/M portfolios, across the columns are the two size groups (Small and
Big) and across the rows are the three B/M groups (Low, Medium, and High). Panel B is the
regression results of six Size-OP portfolios, same as Panel A, across the columns are the two size
groups and across the rows are the three OP groups (Weak, Neutral and Robust). Panel C is the
regression results of six Size-Inv portfolios, across the columns are the two size groups and across
the rows are the three Investment groups (Aggressive, Neutral and Conservative). Numbers in bold
are the t-stats which are significant at 5% confidence level.
Regression: Ri ,t  R f  ai  bi ( RM ,t  R f )  si SMB  hi HML  ri RMW  ci CMA  ei ,t
Panel A: Time-series regressions on six Size-B/M portfolios
Book-to-Market (B/M) ratio
L M H L M H
a t (a )
S -0.2123 -0.0328 0.2391 -1.2440 -0.3040 1.4394
B 0.0339 -0.0369 0.0197 0.2766 -0.2794 0.1673
b t (b)
S 0.9867 1.0036 0.8091 18.0426 37.3893 18.0305
B 1.0715 1.1142 1.0577 30.4708 21.6453 42.6438
s t (s)
S 0.7036 0.7975 0.4312 13.4733 17.5148 5.5548
B 0.2231 0.1980 0.1550 4.6577 3.4748 2.2473
h t (h)
S -0.2137 0.2549 0.3912 -2.0573 4.1878 3.8325
B -0.2152 0.0582 0.6456 -4.9748 0.9063 9.2306
r t (r)
S -0.7327 -0.1351 -0.3846 -5.4871 -2.2599 -3.0270
B -0.1506 -0.0322 0.0110 -2.7377 -0.3318 0.1531
c t (c)
S -0.3047 -0.2155 -0.0974 -2.7655 -2.3334 -0.7657
B -0.2294 -0.0887 -0.1666 -2.3567 -1.1923 -1.4554
S 0.9517 0.9733 0.9084 0.0115 0.0080 0.0125
B 0.9670 0.9591 0.9580 0.0077 0.0088 0.0089
199
Appendix B
Table B.4 Continued
Panel B: Time-series regressions on Size-OP portfolios

Operating Profitability
W N R W N R
a t (a )
S -0.0188 0.0898 -0.0396 -0.3438 1.1708 -0.4349
B -0.0842 0.0989 -0.0640 -0.7461 2.2880 -1.9871
b t (b)
S 0.9812 0.9853 1.0646 81.0285 51.1117 32.7726
B 1.1136 0.9412 1.0298 27.2225 50.5000 97.2971
s t (s)
S 0.8675 0.9675 0.9317 33.4687 20.9395 14.3934
B -0.0693 -0.0541 -0.1339 -1.0857 -1.2950 -4.5316
h t (h)
S -0.1143 0.2669 0.2011 -4.5007 6.3486 3.9980
B 0.2443 0.0392 -0.0708 4.5818 1.0984 -2.7060
r t (r)
S -0.6348 0.2597 0.4475 -18.4610 5.1374 9.5450
B -0.5864 -0.1016 0.3304 -8.4796 -2.7847 12.2607
c t (c)
S 0.0768 -0.0627 -0.1247 1.6662 -1.0611 -1.5351
B -0.2849 0.1389 -0.0839 -3.0221 2.6995 -2.0856
S 0.9945 0.9851 0.9826 0.0040 0.0058 0.0063
B 0.9775 0.9863 0.9903 0.0069 0.0041 0.0033
Panel C: Time-series regressions on Size-Inv portfolios
Investment
A N C A N C
a t (a )
S 0.0071 0.1144 -0.0540 0.1573 2.3620 -0.8730
B 0.0259 -0.0428 0.0831 0.4733 -0.7299 1.1236
b t (b)
S 0.9681 0.9710 1.0880 47.1125 35.9423 81.6853
B 1.0521 0.9912 0.9326 77.8118 48.6599 32.4959
s t (s)
S 0.9702 0.8942 0.8760 30.5318 26.8312 19.7427
B -0.1786 -0.0167 -0.0835 -5.7904 -0.9118 -2.1632
200
Appendix B
Table B.4 Continued
Panel C: Time-series regressions on Size-Inv portfolios

Investment
A N C A N C
h t (h)
S 0.0238 0.1754 -0.0163 0.7234 5.4018 -0.3663
B -0.0658 0.0615 -0.0260 -1.4661 1.3401 -0.4450
r t (r)
S -0.2181 0.1339 -0.2339 -4.7274 3.5684 -5.6108
B -0.0174 0.0573 -0.0001 -0.3659 1.1884 -0.0013
c t (c)
S -0.4264 0.1159 0.3567 -7.7210 2.0770 5.2777
B -0.5734 0.1819 0.6429 -8.1165 3.4479 7.3145
S 0.9907 0.9879 0.9908 0.0048 0.0051 0.0051
B 0.9799 0.9872 0.9764 0.0052 0.0041 0.0054
201
Appendix C
Time-series regression of six Size-B/P portfolios, six Size-

OP portfolios and six Size-Inv portfolios on FF3F Model
(Chinese stock market)
Table C.1 Time-series regression of three sets of portfolios on FF3F Model, Chinese A-share
stock market (July 2010- May 2015, 59 months)
This table presents the time-series regressions results of six Size-B/P portfolios, six Size-OP
portfolios and six Size-Inv portfolios on FF3F Model on Chinese stock market in Panel A, Panel B,
and Panel C respectively. In each panel, the regression coefficients and adjusted R-square are
presented in the left part of the table, the corresponding t-statistics corrected for heteroscedasticity
and autocorrelation using the Newey-West estimator and residual standard error are presented in the
right part. Across the columns of each panel are the two size groups (S and B); across the rows of
Panel A are the three B/M groups (L, M and H)., across the rows of Panel B are the three OP groups
(W, N and R), across the rows of Panel C are the three Investment groups (A, N and C). Numbers in
bold are the t-stats which are significant at 5% confidence level.
Panel A: Time-series regression of six value-weighted Size-B/P portfolios
Size Book-to-Price (B/P) ratio
L M H L M H
a t (a )
S 0.0113 0.0113 0.0112 7.4842 4.5837 6.1412
B 0.0123 0.0092 0.0123 6.2322 3.3146 6.4849
b t (b)
S 0.8979 0.9231 0.9059 32.6701 27.0157 32.2014
B 0.8504 1.0340 0.8424 27.3918 22.4104 22.1472
s t (s)
S 0.9057 0.8874 0.8548 27.9738 15.5660 19.8422
B -0.1441 -0.1149 -0.0931 -3.4693 -1.3686 -2.7201
h t (h)
S -0.4025 -0.3497 -0.0262 -5.0751 -4.3756 -0.4383
B -0.9630 -0.4761 0.6607 -15.1937 -5.5671 7.1594
S 0.9803 0.9782 0.9746 0.0114 0.0120 0.0119
B 0.9623 0.9545 0.9542 0.0122 0.0147 0.0130
203
Appendix C
Table C.1 Continued
Panel B: Time-series regression of six value-weighted Size-OP portfolios

Profitability
W N R W N R
a t (a )
S 0.0038 0.0042 -0.0022 1.3466 0.7791 -0.5303
B 0.0000 0.0027 0.0010 -0.0179 1.5449 0.3059
b t (b )
S 0.9236 0.9733 1.0657 24.3001 17.2826 17.4507
B 1.0967 0.9896 1.1164 19.0453 34.3179 20.3752
s t (s )
S 1.0674 0.8611 0.7540 7.9772 5.1362 5.5995
B 0.4230 0.3508 -0.1688 3.8493 7.9956 -1.5525
h t (h)
S -0.2302 -0.5825 -0.4059 -1.8362 -3.4193 -2.2086
B -0.1523 -0.4188 -0.7060 -1.3302 -4.9840 -6.5449
S 0.9456 0.9158 0.8514 0.0200 0.0261 0.0356
B 0.9520 0.9698 0.9388 0.0174 0.0127 0.0188
Panel C: Time-series regressions of six value-weighted Size-Inv portfolios
Investment
A N C A N C
a t (a )
S -0.0015 0.0024 -0.0002 -0.7269 1.1784 -0.1061
B -0.0038 0.0000 -0.0025 -2.4503 0.0310 -1.3087
b t (b )
S 0.9667 0.9745 0.9521 50.2415 37.7947 34.0437
B 1.0323 1.0393 1.0796 31.6716 37.0603 26.3027
s t (s )
S 1.0170 0.9787 1.1180 16.8657 22.2040 19.8524
B 0.3535 0.3491 0.5370 5.0630 7.1627 8.8230
h t (h)
S -0.3873 -0.4312 -0.2089 -3.2013 -5.7192 -2.3319
B -0.6595 -0.3505 -0.1090 -7.1360 -4.6239 -0.9868
S 0.9508 0.9778 0.9701 0.0197 0.0131 0.0151
B 0.9621 0.9742 0.9595 0.0155 0.0121 0.0160
204
Appendix D
Time-series regressions on single innovation of selected state variables in addition

to excess market return
Table D.1 Time-series regressions on single innovation of state variables (December 2006- May 2015)
This table presents the four time-series regressions on the single innovation of state variables in addition to excess market return, only loadings
on the innovations are reported. The left-hand part of the table presents the loadings and right-hand part is the corresponding t-statistics
corrected for heteroscedasticity and autocorrelation using the Newey-West estimator. Across the columns of each part are the five size quintiles
and across the rows are five B/P ratio quintiles. Numbers in bold indicate statistical significance at 5% confidence level.
Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IDIV IDIV   i ,t
R  R f   i  i ,m ( RM ,t  R f )  i ,ITERM ITERM   i ,t
Regressions: i ,t
Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IDEF IDEF   i ,t
Ri ,t  R f   i  i ,m ( RM ,t  R f )  i ,IRF IRF   i ,t
Size B/P ratio
L 2 3 4 H L 2 3 4 H
Panel A: Regressions on market factor and IDIV
S -19.4121 -23.4954 -21.6904 -20.6291 -20.2875 -2.4626 -2.9855 -3.3198 -2.7667 -2.5375
2 -24.2059 -18.1294 -21.2079 -24.0719 -22.1941 -3.5465 -2.5009 -3.1129 -3.3345 -3.0771
3 -19.6072 -25.9214 -17.4444 -20.3365 -20.8928 -2.8355 -3.7192 -3.1053 -3.8017 -3.6844
4 -16.2248 -17.2302 -16.4696 -17.8200 -20.6227 -2.6344 -2.8928 -3.3502 -3.2671 -4.5735
B -9.5801 -7.8102 -12.3718 -8.8154 -13.2603 -1.7570 -1.8305 -3.1659 -2.1580 -2.4231
205
206
Table D.1 Continued
Size B/P ratio

Appendix D
L 2 3 4 H L 2 3 4 H
Panel B: Regressions on market factor and ITERM
S 2.1979 3.1786 2.8511 1.3541 1.2907 0.4802 0.7913 0.7025 0.3304 0.3179
2 4.3519 3.1261 4.1306 2.5430 0.8026 1.2628 0.9301 1.0870 0.7630 0.2458
3 3.9247 4.4774 2.6859 1.3648 -0.1431 0.9611 1.3439 0.8586 0.4922 -0.0510
4 3.9734 4.1571 2.4330 0.9741 -1.4101 1.3249 1.4546 1.0761 0.4080 -0.6578
B 3.2004 2.5767 0.1241 -0.3270 -4.0864 1.4070 1.6004 0.0775 -0.2406 -1.6547
Panel C: Regressions on market factor and IDEF
S 3.7428 1.5203 2.8085 3.6790 2.4471 0.5640 0.2210 0.4635 0.6228 0.4097
2 1.1381 0.2448 1.4571 3.7144 2.0826 0.1970 0.0427 0.2412 0.6908 0.4492
3 4.1031 1.5287 1.4575 1.5554 2.0615 0.6695 0.2649 0.2756 0.3260 0.5193
4 -0.0922 -0.1408 0.8468 -0.2386 3.9225 -0.0184 -0.0270 0.2104 -0.0533 1.2321
B 3.6214 -0.1257 1.9757 -0.8604 2.0447 0.8670 -0.0392 0.7604 -0.3124 0.5102
Panel D: Regressions on market factor and IRF
S -6.1567 -5.3639 -6.3742 -4.2995 -3.4335 -1.9355 -1.5769 -2.0364 -1.2874 -1.1219
2 -7.6940 -4.5895 -6.7521 -5.6511 -3.2150 -2.6154 -1.3826 -2.3392 -2.0012 -1.3050
3 -6.5573 -5.9295 -5.8271 -3.8025 -3.0693 -2.0189 -2.1871 -2.2262 -1.6033 -1.4693
4 -6.3293 -6.0462 -4.1635 -2.2150 -0.8936 -2.6779 -2.6614 -2.0668 -0.9592 -0.4960
B -4.3878 -2.0402 0.1862 0.1231 1.3394 -2.1198 -1.2005 0.1158 0.0756 0.6946
Appendix E
Derivation process of Merton’s default probability and

distance-to-default
The default probability is the probability that the firm’s assets value is less than the book
value of the firm’s liabilities, which is:
  
Pt  Prob VA,t T  X t VA,t  Prob ln VA,t T   ln  X t  VA,t 
where Pt is the probability of default at time t, V A,t and X t are the market value of the
firm’s assets and the book value of the firm’s liabilities at time t, and VA,t T is the market
value of the firm’s assets due at time T.
Since the value of firm’s assets follows GBM of dVA  VA dt   AVA dW (equation 3.5), the
value of assets at time t is given by:
  A2 
ln VA,t T   ln VA,t       T   A T  t T
 2 
W t  T   W t 
where  t T = ,  t T N 0,1 , µ is the drift rate which is the expected
T
return on the firm’s asset, and ε is the random component of the firm’s return which is
normally distributed assumed by BS model.
Then the probability of default above can be rewritten as follows:
  2  
Pt  Prob  ln VA,t      A  T   A T  t T  ln  X t  
  2  
  VA,t    A2  
 ln  
    T 
 Prob    t
X   2 
  t T 
 A T 
 
 
According to the normal distribution of ε, the probability of default Pt can be defined in

terms of the cumulative normal distribution:
207
Appendix E
  VA,t   1 2 
 ln        A T 
P=N    Xt   2  
t
 A T 
 
 
which is the Merton’s probability of default.
208
Appendix F
Time-series and cross-sectional regressions of 18 Size-
B/P-DR sorted portfolios on FF3F model
Table F.1 Time-series regressions of 18 Size-B/P-O-score sorted portfolios on FF3F Model,

Chinese stock market (July 2005 to May 2015)
This table reports the time-series regression results for 18 Size-B/P ratio-O-score on FF3F Model.
The stocks are divided into two size groups, three B/P groups, and three groups based on their O-
score, seperately. The intersection of these groups forms 18 portfolios. The left part of the table is
the coefficients of time-series regressions and adjusted R-square, the right part is the corresponding
t-stats corrected for heteroscedasticity and autocorrelation using Newey-West estimator and residual
standard error, and the numbers in bold are significant at 5% confidence level.
Small size
O-score B/P ratio
L M H L M H
a t (a )
O1 0.0003 -0.0015 -0.0042 0.0950 -0.6211 -1.4931
O2 -0.0072 -0.0019 -0.0019 -2.0880 -0.8494 -0.6544
O3 -0.0056 -0.0011 0.0000 -1.8313 -0.3788 -0.0104
b t (b)
O1 0.9738 0.9995 0.9741 18.6385 19.9970 21.0746
O2 1.0262 0.9931 1.0063 18.9585 23.7498 17.0678
O3 0.9945 0.9981 0.9964 18.7994 18.4143 16.3521
s t (s)
O1 1.3451 1.3748 1.5414 11.4333 12.7305 14.9231
O2 1.4930 1.3699 1.3850 13.5019 15.0567 11.7254
O3 1.5798 1.4035 1.3333 15.4323 12.6156 11.2084
h t (h)
O1 -0.1961 0.1113 0.5427 -1.3753 0.8522 3.9638
O2 0.0477 0.1810 0.4657 0.3208 1.4722 3.4695
O3 0.1713 0.3048 0.4942 1.2575 2.3177 3.4456
O1 0.8677 0.9121 0.9038 0.0404 0.0329 0.0355
O2 0.9000 0.9179 0.9031 0.0368 0.0315 0.0353
O3 0.9072 0.9058 0.8986 0.0352 0.0344 0.0356
209
Appendix F
Table F.1 Continued
Big size
O-score B/P ratio
L M H L M H
a t (a )
O1 -0.0017 -0.0029 -0.0046 -0.7289 -1.0484 -1.7760
O2 -0.0036 -0.0005 -0.0048 -1.2359 -0.1598 -1.8854
O3 -0.0060 -0.0063 -0.0007 -1.8987 -2.1111 -0.2422
b t (b)
O1 0.9549 1.0330 1.0255 18.6257 20.9052 18.9205
O2 1.0119 1.0910 1.0489 18.3348 19.0292 16.7194
O3 1.0332 1.1153 1.0781 15.0781 16.2983 22.2291
s t (s)
O1 0.4950 0.7539 0.8142 5.1068 6.2125 8.9078
O2 0.7978 0.6842 0.7971 7.4561 6.0882 7.4711
O3 0.8567 0.8561 0.7651 7.0660 6.7420 7.2620
h t (h)
O1 -0.6448 -0.0470 0.4973 -5.0173 -0.3281 4.4956
O2 -0.3376 0.0427 0.6607 -2.3818 0.3042 4.6627
O3 -0.2528 0.1370 0.8074 -1.6469 0.9250 6.4202
O1 0.8678 0.8800 0.9066 0.0345 0.0361 0.0323
O2 0.8750 0.8936 0.9081 0.0364 0.0353 0.0331
O3 0.8542 0.8781 0.9109 0.0409 0.0398 0.0337
210
Appendix F
Table F.2 Time-series regressions of 18 Size-B/P-DLI sorted portfolios on FF3F Model,

Chinese stock market (July 2005 to May 2015)
This table reports the time-series regression results of 18 Size-B/P-DLI sorted portfolios on FF3F
Model. The stocks are divided into two size groups based on the breakpoint of median market
capitalization, and the breakpoints for the three B/P groups are the top 30%, median 40% and
bottom 30% of B/P ratio, similarly as B/P groups, the stocks are divided into three groups based on
their O-score (O1, O2 and O3, represent the low, medium and high O-score separately). The
intersection of two size groups, three B/P groups and O-score groups form 18 portfolios. The left
part of the table is the coefficients of time-series regressions and adjusted R-square, the right part is
the corresponding t-stats corrected for heteroscedasticity and autocorrelation using Newey-West
estimator and residual standard error, and the numbers in bold are significant at 5% confidence level.
Small size
DLI B/P ratio
L M H L M H
a t (a )
D1 -0.0025 0.0002 -0.0031 -0.9101 0.0587 -0.9167
D2 -0.0081 -0.0019 -0.0005 -2.8124 -0.8704 -0.2072
D3 0.0014 0.0004 -0.0013 0.3786 0.1655 -0.5089
b t (b)
D1 0.9817 0.9758 0.9545 19.5959 21.2488 23.9619
D2 1.0116 1.0041 1.0103 20.3742 22.2973 20.1025
D3 1.0067 1.0209 1.0050 19.4264 25.5842 16.4963
s t (s)
D1 1.4510 1.4549 1.6158 17.1267 12.7620 13.5801
D2 1.6063 1.4122 1.4775 14.7615 16.0700 12.1172
D3 1.4346 1.3751 1.3206 14.0215 14.3170 12.4977
h t (h)
D1 -0.2634 -0.1172 0.4312 -1.9823 -0.8851 2.8019
D2 -0.0145 0.0316 0.3284 -0.1077 0.2800 2.3027
D3 0.1395 0.2050 0.4360 1.0807 1.7397 3.3278
D1 0.9173 0.9073 0.8990 0.0321 0.0339 0.0364
D2 0.9211 0.9286 0.9087 0.0328 0.0297 0.0347
D3 0.8847 0.9205 0.9090 0.0389 0.0317 0.0336
211
Appendix F
Table F.2 Continued
Big size
DLI B/P ratio
L M H L M H
a t (a )
D1 -0.0001 -0.0032 -0.0051 -0.0287 -1.2382 -1.6917
D2 -0.0054 -0.0022 -0.0022 -1.9525 -0.8175 -0.8457
D3 -0.0027 -0.0051 -0.0039 -0.8710 -1.6936 -1.6293
b t (b)
D1 0.9290 1.0053 0.9359 20.0667 27.0452 17.9607
D2 1.0405 1.0926 1.0638 15.5261 20.9450 20.9343
D3 1.0942 1.1301 1.0846 22.2336 17.9663 18.8849
s t (s)
D1 0.6290 0.8346 0.9219 5.9958 7.5760 8.1997
D2 0.7953 0.7991 0.7913 6.3616 7.6778 8.4618
D3 0.6881 0.7427 0.7786 6.7352 5.4622 7.5786
h t (h)
D1 -0.6961 -0.2435 0.4531 -5.5203 -1.6542 3.5546
D2 -0.4687 -0.1322 0.4000 -3.2566 -0.8330 3.3928
D3 -0.2559 0.0840 0.7448 -1.7169 0.6183 5.6516
D1 0.8559 0.8798 0.8687 0.0362 0.0356 0.0366
D2 0.8724 0.8993 0.9105 0.0379 0.0346 0.0323
D3 0.8781 0.8830 0.9203 0.0380 0.0387 0.0317
212
Appendix G
List of publications and conferences
Journal
Jiao, W., and Lilti, J.J. (2017), Whether profitability and investment factors have additional
explanatory power comparing with Fama-French Three-Factor Model: empirical evidence
on Chinese A-share stock market. China Finance and Economic Review, 5(1), 7.
Conference
Jiao,W., and Zhang, J. (2016), Do the innovations of predictive variables explain Chinese
stock market? The 13th Edition of Augustin Cournot Doctoral Days (ACDD), April 21st and
22nd of 2016, Strasbourg, France.
Jiao, W., and Lilti, J.J. (2017), Exploring Fama-French Five-Factor Model on Chinese A-
share stock market. The 34th International Conference of the French Finance Association,
May 31st, June 1st and 2nd of 2017, Valence, France.
213

JIAO Wenting

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

JIAO Wenting

Transféré par

Droits d'auteur :

Formats disponibles

Exploring Risk Factors on Chinese A Share Stock

Market - in the Frame of Fama - French Factor Model

To cite this version:

HAL Id: tel-01685829

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

THÈSE / UNIVERSITÉ DE RENNES 1

École doctorale EDGE

Thèse soutenue à Rennes

Factors on Chinese Khalid EL BADRAOUI

A-Share Stock Market Université Ibn Zohr (Maroc) / rapporteur

To everyone who I love

List of Tables ....................................................................................................................... vii

List of Figures ...................................................................................................................... xi

Abbreviations ..................................................................................................................... xiii

Résuméde la thèse en français ........................................................................................ xvii

Chapter 1 Fama-French Five-Factor Model vs. Fama-French Three-Factor Model in

Chapter 2 Fama-French Factors and Innovations in State Variables on Chinese A-

2.1 Economic explanation of Fama-French factors in the context of ICAPM ............ 70

3.3 Financial distress risk and equity returns ............................................................. 135

General conclusions .......................................................................................................... 169

Bibliography ...................................................................................................................... 173

Appendix D Time-series regressions on single innovation of selected state variables in

Appendix E Derivation process of Merton’s default probability and distance-to-default 207

Appendix F Time-series and cross-sectional regressions of 18 Size-B/P-DR sorted

Appendix G List of publications and conferences ............................................................ 213

Table 1.1 Summary of Chinese stock market 1990-2014 .................................................... 34

Table 1.14 Time-series regressions of six value-weighted Size-B/P portfolios, Size-OP

Figure 1.1 SHASHR Index and SZASHR Index (1992-2015)............................................. 33

3M T-Bill Rate: Three-Month Treasury Bill Rate

ANNA: Artificial Nerve Network Analysis

B/M: Book-to-market equity

BSM: Black-Scholes and Merton

CDS: Credit default swaps

Chinese A-share: CNAS

Cross-sectional regression: CSR

CSRC: China Securities Regulatory Commission

DEF: Default spread

DIV: Dividend yield

DLI: Default likelihood indicator

DP: Default (distress) probability

DR: Distress (default) risk

DRF: Distress risk factor

EDF: Expected default frequency

FF3F Model: Fama-French Three-Factor Model

FF5F Model: Fama-French Five-Factor Model

GBM: Geometric Brownian Motion

GEB: Growth Enterprise Board

GDP: Gross Domestic Product

GNP: Gross National Product

ICC: Implied cost of capital

IDEF: Innovations of default spread

IDIV: Innovations of dividend yield

IHML: Innovations of HML

IRF: Innovations of one-month deposit rate

ISMB: Innovations of SMB

ITERM: Innovations of term spread

LTD: Long-term debts

MDA: Multiple discriminant analysis

OP: Operating profitability

PT: Particular transfer

RF: Risk-free Rate (one-month deposit rate)

RMB: RenMinBi (Chinese ‘yuan’)

SDF: Stochastic Discount Factor

SHASHR Index: Shanghai Stock Exchange A-Share Index